🚀 Start here
Free is not the same as weak. Coding agents and models available on a $0 plan.
Every model name here is read back from the vendor's own API or pricing page twice a week, and the quota that comes with it is in the agents table.
Free API access
One key, many free models
No account at all
A trial that asks for no card
Claude Code on a free lane
Codex CLI on a free lane
- OpenRouter (free models)
codex -p openrouter-free - Kilo Code
codex -p kilo-code - AIHubMix (free models)
codex -p aihubmix - Any lane of the LiteLLM config
codex -p litellm
Every answer is the registry's own ranking, explained in CONTRIBUTING, read from the top — of its section, or of every section for no account at all, Claude Code and Codex CLI — so a row that stops verifying leaves these cards on the same run it leaves the list.
No account at all? Paste this.
Kilo Code answers in the terminal you already have open — a rate-limited lane to prove this page is live, not a setup to write code on.
curl -s https://api.kilo.ai/api/gateway/chat/completions \
-H 'Content-Type: application/json' \
-d '{"model":"openrouter/free","messages":[{"role":"user","content":"2+2? MAKE NO MISTAKES."}]}'
Use free IDs without a key; metered IDs require sign-in. kilo-auto/free and openrouter/free choose a free model automatically. LiquidAI advises against agentic coding with lfm-2.5-2.6b.
🤖 Coding agents & CLIs
23 live · not one of them asks for a card · sorted by how much work you can get done for free
| Tool | What you get | Free models | Limits, in the vendor's words | Verified |
|---|---|---|---|---|
| GitHub Copilot Free | Free Copilot plan for individual developers in VS Code, JetBrains, Visual Studio and CLI; completions, limited chat and agent usage | — |
2,000 completions/month per account …Chat and agent usage: Monthly credits allowance: amount not published per account Chat and agent features spend GitHub AI Credits, whose Free amount is unpublished. Model selection is automatic on Copilot Free and Student. Free is available only to individual developers who do not receive Copilot through an organization or enterprise. |
|
| opencode | Open-source coding agent whose opencode Zen gateway prices a rotating set of models at zero — inside OpenCode only, no sign-in; any provider via BYOK |
Free model usage: Usage allowance: amount not published (period not published); scope not published …Free models work only inside OpenCode; other clients receive `403 FreeTierError: OpenCode's free tier can only be used from within OpenCode`. Use `opencode/<model-id>` and select the Free variant. Offers rotate; each named free offer is "available on OpenCode for a limited time". Collected free-model data may improve models; NVIDIA offers are trial-only and exclude personal or confidential data. Flint 40B Free allows training and retains data for 30 days. Beam Free's provider does not train on inputs and retains them for 30 days; its listed territorial exclusions include Russia. |
||
| Kilo Code | Open-source coding agent with an Auto Free mode and a keyless OpenAI-compatible gateway, alongside BYOK and local models | 200 requests/hour per IP The free-model lineup rotates. Auto Free routes over the free models in the catalog's autoRouting list. Other models use pay-as-you-go credits or a Kilo Pass subscription. | ||
| Kiro | Perpetual free tier of AWS's spec-driven agentic IDE (successor to Amazon Q Developer) with Claude and open-weight models |
50 credits/month per account …Requires social login or AWS Builder ID. Credits do not roll over. Free requests are served from the US and AWS GovCloud is excluded. The docs' Free column determines model eligibility; newer Sonnet, Opus, Haiku and GPT tiers marked paid are outside the Free allowance. |
||
| Google Antigravity | Google's agent-first IDE and CLI, with a $0 Individual plan serving Gemini and third-party agent models |
Gemini agent allowance: Weekly usage allowance: amount not published per account …Claude and GPT agent allowance: Weekly usage allowance: amount not published per account Tab: Unmetered completions per account Command: Unmetered requests per account Follow the Free & Google AI Plus column of the model table; Enterprise-only Gemini models are outside this offer. Older third-party models in that column are scheduled for removal on November 2, 2026. Claude 5.5 and Gemini 4 Argon require paid plans. The plans overview instead lists third-party access under Ultra, so those sources disagree. Gemini and third-party models have separate weekly allowances. Baseline limits depend on capacity and abuse prevention; no numerical weekly budget is published. |
||
| Freebuff | Ad-funded coding agent for CLI, desktop, web, cloud and chat, with daily regional allowances and conditional free-session access |
Session-based limited mode: 6 one-hour sessions per day per account …Available models depend on the app and access level. “Promotional prices are temporary, and any price may change.” “Freebuff collects prompts, messages, code, files, repository data, and agent traces when you use features that need them”. Check the balance and picker in your app. Daily allowance — the United States, Canada, the United Kingdom, Australia, New Zealand, Ireland, Norway, Sweden, Denmark, Finland, the Netherlands, Austria, Luxembourg, Iceland, Germany, France, Spain, Italy, Portugal, Belgium, Switzerland, Liechtenstein, Malta, South Korea: 30 hours per day; Other countries: 5 hours per day; VPN or proxy: 4 hours per day. Check your account's picker for current model hour prices. Solar Pro 4: shared daily hours; available in limited mode; Solar Mini 4: shared daily hours; available in limited mode; MiMo V2.6 Flash Fast: shared daily hours; available in limited mode; GLM 5.3 Flash: shared daily hours; available in limited mode; DeepSeek V4.1 Flash 10/17: shared daily hours; available in limited mode; GPT-6 Luna: shared daily hours; full access only; MiMo 2.6 Pro: shared daily hours; full access only; DeepSeek V4.1 Flash Fast 10/17: shared daily hours; full access only; GPT-6.1 Terra: shared daily hours; full access only; GPT-6.1 Sol: paid plan; Gemini 4 Argon: paid plan; Gemini 3.8 Flash: paid plan; Muse Spark 1.3: shared daily hours; full access only; Gemini 3.1 Flash Lite: specialist tasks only. Your balance is measured in hours. Gemini 3.8 Flash: “PAID-ONLY row” (source: https://raw.githubusercontent.com/CodebuffAI/codebuff/main/common/src/constants/freebuff-models.ts). GPT-6.1 Sol, Gemini 4 Argon: The public Free-plan comparison requires a paid plan; paid plans allow one one-hour session per day. (source: https://freebuff.com/plans). Gemini 3.1 Flash Lite handles specialist tasks such as file finding and research. Muse Spark 1.3: Credit-funded at full access; a paid plan is required at limited access. (source: https://raw.githubusercontent.com/CodebuffAI/codebuff/main/common/src/constants/freebuff-models.ts). GPT-6.1 Terra: “Open on full access, a paid plan on limited access. One session a day on every plan.” (source: https://raw.githubusercontent.com/CodebuffAI/codebuff/main/common/src/constants/freebuff-sol-promo.ts). MiMo V2.6 Flash Fast is the default on CLI, Desktop, Web, and Cloud. Daily hours refill at midnight in your local timezone and unused daily hours do not carry over. Older app versions show the same balance as Freebucks, five to an hour. (source: https://freebuff.com/llms.txt). |
||
| Cline | Open-source coding agent for VS Code, JetBrains and the terminal; signing in to its own Cline provider unlocks a rotating set of free models, each with a daily allowance, beside pay-as-you-go credits, the $9.99 ClinePass plan and BYOK |
Free model promotions: Daily usage allowance: amount not published per account per model …Free promotions require a Cline account, with Google, GitHub or email signup and no card. The daily per-model scope is also reported by the client stop message: `You've reached today's free usage limit for this model`. The picker's free list rotates. These models work only in the Cline IDE extension and CLI, not through the Cline API. Free usage may improve model performance and quality. |
||
| Google Jules | Free tier of Google's async cloud coding agent, on Gemini models; connects to GitHub repos and works autonomously |
gemini-2.5
|
15 tasks/rolling day; 3 tasks at once per account …The limits table still names Gemini 2.5 Pro for Free, while the 2026-01-30 changelog says Gemini 3 Flash became the base model for all tiers. Google AI upgrades require an individual @gmail.com account; family-plan task limits are not pooled. Google may adjust limits and features. |
|
| Cursor (Free) | Free coding plan of the Cursor AI IDE, with limited Agent and Chat usage | — |
Agent usage: Usage allowance: amount not published (period not published) per account …Free includes limited Agent, Chat and Tab usage with Cursor's first-party models; numerical allowances are unpublished and third-party models are not included. The pricing FAQ describes weekly refreshes across tiers, while detailed docs explicitly assign weekly pools to paid X subscriptions and monthly pools to legacy Cursor plans. Those descriptions do not establish a consistent Free reset; check your account. |
|
| Devin Desktop (formerly Windsurf) | Free plan of Cognition's desktop coding agent — the IDE that shipped as Windsurf |
swe-1.6
|
Agent usage: Daily usage allowance: amount not published; Weekly usage allowance: amount not published per account …Tab: Unmetered completions per account Inline edits: Unmetered interactions per account Windsurf is now Devin Desktop. Free has limited model availability; the model docs name SWE-1.6 for Free. Daily and weekly agent windows both apply. At a cap, Free must wait for a reset; buying extra usage is a paid-plan feature. A separate SWE-2 promotion appears on the Pro card and does not establish eligibility for the Free plan. |
|
| Autohand Code | Open-source coding agent for the terminal, VS Code, JetBrains and the desktop whose Free plan runs Autohand's own latency-first coding model with no card, using recurring hosted-response credits |
fantail
|
15 credits/rolling 5 hours; 100 credits/month; 10 requests/minute; 1,000 input tokens/minute; 1,160 output tokens/minute per account …Ongoing Free access is intended for evaluation and light use. It requires a linked GitHub account at least 30 days old, an invite, an Autohand grant, or an optional card on file that is not charged. One credit counts one successful hosted model response; BYOK and local models do not consume these credits. All credit, request and token limits are shared across product surfaces. Additional hosted capacity and models need a paid plan. |
|
| MiniMax Code | MiniMax's coding agent — a desktop app for macOS and Windows, with an open-source CLI beside it — whose daily check-in in the app earns free points to spend on its tasks, coding among them | — |
Standard check-in: 400 points/day per account (Points valid for 30 days) …Check-in days 4 and 7: 1,000 points/day per account (Points valid for 30 days) The higher day-4 and day-7 check-ins replace the normal daily award rather than add to it. Check-in points cover MiniMax Code task scenarios; their spending value follows the product UI. The desktop check-in names no payment step. Global and mainland accounts use separate sites. The CLI accepts account credits or BYOK, but no source establishes check-in-point eligibility there. Users must be at least 16. |
|
| Codex (ChatGPT Free) | Codex desktop access on a $0 ChatGPT Free plan, subject to rollout |
Free plan: Usage allowance: amount not published (period not published) per account …The Free card names GPT-6 Luna. The Plus card explicitly includes web, CLI, IDE extension and iOS access; those surfaces are not specified on the Free card. The plan cards publish no numerical Free allowance. |
||
| CodeGPT | VS Code / JetBrains coding agent whose $0 plan includes model usage rather than only BYOK — a small daily allowance on its own Economy models, plus BYOK across 15+ providers and local models (Ollama, LM Studio) beside it | — |
Economy models: 10 interactions/day per account …Premium models: $1 once per account No credit card is required. The Economy tier is described as free forever, but its routed model names are unpublished. Unlimited Economy use requires the paid Professional plan. The premium welcome credit is a separate one-time mode. |
|
| Qoder | Alibaba's coding apps with an ongoing Free basic-model allowance, a free Qwen promotion, a 300-credit Pro trial, and daily promotional credits to claim |
Pro trial: 300 credits once per account (Valid for 2 weeks) …Claimed promotional credit: 100 credits/day per account (Each claim valid for 30 days) Pro trial completion and next edits: Unmetered completions per account (Valid for 2 weeks) Qwen3.8-Flash promotion: Daily requests allowance: amount not published per account per model; for `Qwen3.8-Flash` The Free plan continues on unnamed basic models with unpublished daily and monthly limits after credits run out. The Qwen3.8-Flash promotion includes individual Free accounts with zero Credits, requires no claim, and excludes Teams and Enterprise. Daily free uses are limited, with more for paid accounts; after the allowance, wait until the next day or spend Credits. Free uses spend no Credits; its end will be announced in advance. The DeepSeek-Flash promotion ended on 2026-11-01 at 00:00 Singapore time. The separate 14-day Pro trial includes 300 credits and unmetered completions and next edits. Daily promotional credits require a manual claim in the international desktop app; "End time: To be announced". Efficient-tier promotional usage requires a paid plan. |
||
| JetBrains AI (AI Free) | AI Free in JetBrains IDEs — unlimited code completion on JetBrains' own model and 3 AI Credits ($3) of cloud models every 30 days for chat and agents |
mellum
|
AI Free: 3 credits/30 days per account …Unmetered completions per account The credit window resets every 30 days from first use; AI Credits are not a guaranteed token count. Mellum completion is a separate unmetered free mode. An active eligible IDE licence is required, including educational, open-source, trial or supported non-commercial licences. IntelliJ without Ultimate, PyCharm without Pro and Android Studio are excluded. Only the optional AI Pro trial asks for a card. Serviceable territories exclude mainland China and other provider-restricted regions. |
|
| Baidu Comate (文心快码) | Baidu's coding assistant for VS Code, JetBrains, Visual Studio and Xcode, with its own IDE and CLI; the free Personal Standard plan includes code completion and a one-time 5,000-point grant valid for seven days | — |
5,000 points once per account (Valid for 7 days) …Personal Standard completion: Completions allowance: amount not published (period not published) per account Personal Standard completion is free, with no published amount or model name. The points grant is introductory. Auto-Free fallback is a paid-plan benefit; Personal Standard does not receive the paid hourly token fallback. |
|
| TRAE (TraeCode) | Free plan of TRAE's AI IDE, TraeCode: Auto mode only, a one-time Basic usage allowance and limited autocompletion | — |
Basic usage: One-time USD allowance: amount varies per account …Autocompletion (where listed for Free): 5,000 completions/month per account TRAEWORK (where listed for Free): 2 tasks at once per account Free uses Auto mode, whose routed models are unpublished. The Basic dollar balance is a one-time grant; its amount differs between served versions of the pricing page. Measurements on 2026-09-16 found $1 and $3 on different networks; a runner read $1 on 2026-09-27. The payload also lists 1,000 advanced-model requests without stating their period. Published completion and concurrent-task figures apply where that Free table is served. Other versions describe limited usage and autocomplete without numbers. |
|
| CodeBuddy (Tencent) |
Tencent's VS Code / JetBrains / CLI coding agent whose Free plan carries 100 credits a month, a daily activity bonus of 30 and 250 welcome credits …Tencent's VS Code / JetBrains / CLI coding agent whose Free plan carries 100 credits a month, a daily activity bonus of 30 and 250 welcome credits, with every model open while the promotion runs — its site and the agent's own endpoint unreachable from the US, India and Russia |
— |
Base Free plan: 100 credits/month per account …Promotional daily bonus: 30 credits/day per account Standard completion allowance: 5,000 completions/month per account Current promotion: Unmetered completions per account Signup credit: 250 credits once per account (Valid for 14 days) The current promotion makes completions unmetered and enables auto routing across all available models. "Standard quotas resume when the promotion ends" and "The end date of this promotion will be announced separately". Monthly base and bonus credits expire in their month without rollover. Free needs no card; the Pro trial does. Models are unnamed. Users must be at least 18. A 2026-09-25 DNS measurement found the international website, login and extension endpoint unavailable from US, Indian and Russian networks; the docs are also served on workbuddy.ai. |
|
| Qoder CN (formerly Lingma) | Alibaba Cloud's coding apps for China, formerly Tongyi Lingma, with a free personal-model promotion alongside a two-week 300-credit trial and 100 credits a day to claim in the desktop app |
Pro trial: 300 credits once per account (Valid for 2 weeks) …Claimed promotional credit: 100 credits/day per account (Each claim valid for 30 days) Basic-model fallback: Completions allowance: amount not published (period not published) per account Qwen3.8 Flash promotion: Daily requests allowance: amount not published per account per model; for `Qwen3.8-Flash` The model promotion is open to personal users, including free accounts with zero credits; it needs no paid subscription or claim and works across the CN apps, including CLI. Daily free uses are limited, with more for paid accounts; after the allowance, wait until the next day or spend Credits. Its end is "结束时间将提前在本页公告" (to be announced in advance). The separate daily credit promotion must be claimed in the desktop app; its end is "结束时间：另行公告" (to be announced). Each claim window runs from 10:00 to the next day at 10:00 Beijing time (UTC+8); unclaimed credit does not accumulate. After trial credit runs out, a basic model still provides limited completion. Signup requires an Alibaba Cloud account bound to a phone number; real-name verification is needed for purchases. Qoder and Qoder CN are "separate product systems"; their accounts, tasks and data are not interoperable. |
||
| CodeRabbit | Code reviews for local Git changes through an installed CLI or VS Code extension, with an ongoing Free plan alongside a separate introductory trial | — |
Free CLI reviews: 3 reviews/rolling hour per person …Free IDE reviews: 3 reviews/rolling hour per person Sign-in is required. New organizations start with a separate "14-day Advanced trial". Free PR access provides summaries; open-source PR reviews have a separate allowance. Proprietary code is not used for model training, but open-source code may train the service. Review-personalization data is stored unless you opt out of storage. |
|
| cubic | AI reviews for existing repositories, with an ongoing Free PR allowance and an included model for interactive local CLI reviews during alpha | — |
Free PR reviews: 20 reviews/month per organization …Interactive CLI during alpha: Unmetered reviews; scope not published Sign in with a cubic account. A separate "7-day free trial" requires no card. Free PR reviews reset on the organization date shown in Billing. Interactive CLI usage is unlimited during alpha, subject to fair-access rate limits; its end is not published. CI/API-key reviews require a seat on an active subscription. Terms permit cubic to train and improve its own AI systems using submitted content; third-party model providers may not train their own models on it. The CLI docs say "Attribution tracking is enabled by default" and document how to disable it. |
|
| Greptile | AI pull-request reviews on existing repositories with recurring Starter credits; installed CLI reviews use a separate no-card trial or a paid plan | — |
Starter: 50 credits/month; scope not published …Startup grant on approval: 500 credits/month; scope not published (Application required; pre-Series A, under $2M revenue in the past 12 months; grant lasts 6 months) Starter permits one active developer and unlimited repositories. Its credits do not cover CLI reviews; new CLI organizations receive a separate "14-day free trial" with no payment method required. De-identified customer data can be used for AI training; organization admins can turn this off in Data & Privacy. Startup-grant activation and payment requirements are not specified on the pricing page. |
🔌 LLM APIs with free tier
33 live · 31 of them ask for no card · sorted by how much work you can get done for free
| Tool | What you get | Free models | Limits, in the vendor's words | Verified |
|---|---|---|---|---|
| Google AI Studio (Gemini API) | Free tier on the Gemini API, priced model by model rather than as one account quota |
Usage allowance: see your account for values (period not published) per project per model …Free eligibility is priced per model. The current Flash, Flash-Lite and Gemma rows have a free input/output column; Gemini 3.1 Pro Preview and Omni Flash do not. Access to the 2.5 models is restricted to projects that previously used them. Actual rate limits are visible after signing in to AI Studio. Free-tier prompts and responses may be used to improve Google products; paid usage has different data-use terms. Context caching is free on the Flash models and unavailable on the Flash-Lite ones. |
||
| Groq | Fast inference against a free plan Groq publishes as a per-model rate table |
Coding models: 30 requests/minute; 1,000 requests/day; 8,000 tokens/minute; 200,000 tokens/day per organization per model; for `openai/gpt-oss-120b` …Coding models: 30 requests/minute; 1,000 requests/day; 8,000 tokens/minute; 200,000 tokens/day per organization per model; for `openai/gpt-oss-120b`, `openai/gpt-oss-20b`, `qwen/qwen3.8-27b` These are the Free Plan limits for the coding models, shared by keys in the same organization. Classifiers, guardrails, speech and voice models have separate limits; their API examples do not establish free coding-model eligibility. Groq calls the table "a high level summary and there may be exceptions"; the account limits page gives the exact values. |
||
| NVIDIA NIM (build.nvidia.com) | Free endpoints for the models build.nvidia.com marks "Free Endpoint", open-weight models from several labs among them, called with a free NVIDIA Developer Program key at integrate.api.nvidia.com/v1 (OpenAI-compatible) |
glm-5.3-flashnemotron-3-ultralaguna-xs-2.1muse-glimmer-30bnemotron-3.5-lightninggemma-4-31bgpt-oss-20bnemotron-3-nano-omni
+6 more
diffusiongemmallama-3.2-90b-visionllama-3.2-11b-visiondeepseek-v4.1-flashnemotron-3.5-super-vl-120b-a12bstep-5-preview-600b-a27b |
Published ceiling: 40 requests/minute; 10,000 requests/day per account …No card, but an API key requires an NVIDIA Developer Program account and phone verification in a supported country. Published ceilings can vary by model and traffic can cause throttling. Vendor forum reports describe new personal keys that can list models but receive 404 on chat calls; an issued key alone does not establish working inference. Model pages identify available and deprecated endpoints; copy current IDs from the catalog. |
|
| Z.ai (Zhipu GLM) | GLM Flash models free on the API, vision included (OpenAI-compatible at api.z.ai/api/paas/v4) |
Free Flash models: Usage allowance: amount not published (period not published) per account …GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash have free input and output prices. The flagship rows' limited-time-free cached-storage column does not make their input or output free. No numerical rate or usage allowance is published. |
||
| Cloudflare Workers AI | Serverless model inference on Cloudflare, with an OpenAI-compatible API and a daily free allowance | — |
10,000 neurons/day per account. All limits reset daily at 00:00 UTC. …Free text generation: 300 requests/minute per account Past the free allowance "further operations will fail with an error". Some models require a paid billing method through the Workers Paid plan or prepaid AI Gateway credits even if neurons remain; the pricing page identifies them. The free text-generation rate does not apply to models requiring paid access. |
|
| LLM7.io | OpenAI-compatible API with a free dashboard token and a recurring allowance of 100,000 input plus output tokens per day on eligible turbo models | — |
100,000 tokens/24 hours; 1 request/second; 60 requests/minute; 250 requests/hour per key …Get a free token from dash.llm7.io; no anonymous plan is listed. Free quotas may be reduced without notice. Turbo IDs can be used by free tokens, but turbo alone does not establish eligibility: usage_based_only denotes paid usage. The operator names no upstream model provider. |
|
| Hetzner Inference API | OpenAI-compatible API on Hetzner's own EU hardware, free for as long as the experiment runs |
4,000,000 input tokens/60 seconds; 100,000 output tokens/60 seconds; 10 requests/60 seconds per key …Free while the Inference API remains experimental; Hetzner says it will email advance notice of a change. Per-key request and token windows both apply, with HTTP 429 at a cap. No daily, monthly or lifetime cap or end date is published. The service is offered as is for experimental use, without guaranteed availability or backups. Minting a token needs a Hetzner account; payment verification can include a card charge or other routes. |
||
| Alibaba Cloud Model Studio (DashScope, international) | Model-specific free token grants in Singapore, typically one million tokens valid for 90 days; some models retain only previously issued allowances; OpenAI-compatible |
qwen3.8-maxqwen3.8-flashqwen3.8-2.4t-a95bqwen3.8-27bdeepseek-v4-prodeepseek-v4-flashdeepseek-v4.1-flashglm-5.3
+20 more
glm-5.2qwen3.7-maxqwen3.7-plusqwen3.7-flashqwen3.6-plusqwen3.6-flashqwen3.6-35b-a3bqwen3.6-27bqwen3.5-plusqwen3.5-flashqwen3.5-397b-a17bqwen3.5-122b-a10bqwen3.5-35b-a3bqwen3.5-27bqwen3-coder-flashqwen-maxqwen-plusqwen-flashqwen3-vl-plusglm-5.1 |
Typical model token grant: 1,000,000 tokens once per account per model (Where granted; valid for 90 days from activation, model release or approval …Typical model token grant: 1,000,000 tokens once per account per model (Where granted; valid for 90 days from activation, model release or approval, whichever is later) Removed model-grant campaigns: Usage allowance: see your account for values (period not published) per account per model (only previously issued quotas, until expiry or exhaustion; no new grants) Separate OAuth allowance: 2,000 requests/day; scope not published (separate from API-key token grants) Free token grants apply in Singapore (international), independently per model and dated snapshot. Current pricing tables decide new-grant eligibility. Some older models and snapshots no longer receive new grants; previously issued quotas remain usable until expiry or exhaustion. Check the console for each model's remaining quota. Kimi and glm-5.2-fast-preview have no API-key grant. Account information must be completed before activation. After the grant, usage is automatically billed pay-as-you-go unless Free Quota Only is enabled separately for the model; that switch is disabled by default. OAuth has a separate allowance; its eligible models and reset time are not established here. |
|
| IteraCompute | OpenAI-compatible API with recurring free usage on a self-hosted open model |
qwen3.8-27buntil 2026-11-24
|
Qwen Free tier: 30,000,000 tokens/month per account; for `qwen/qwen3.8-27b`. Allowances reset on the first of each month at 00:00 UTC (08:00 Beijing …Qwen Free tier: 30,000,000 tokens/month per account; for `qwen/qwen3.8-27b`. Allowances reset on the first of each month at 00:00 UTC (08:00 Beijing time); unused tokens do not roll over. Qwen Free tier: 160,000 tokens/minute per account; for `qwen/qwen3.8-27b` Register and verify an email, or sign in with Google. Input and output, including reasoning, count across all account keys. "Context includes input and requested output; output is capped at 8,192 tokens." Free total context is 128K. Requests reserve maximum usage, then settle actual usage. Exhausted quota returns 429 without automatic paid overage. No recharge is required; optional cumulative $10/$50 top-ups unlock larger allowances. Recharge thresholds use completed principal minus refunds, excluding grants and fees. Spending the balance keeps the tier; refunds may lower it. Shared capacity may queue; honor Retry-After. No SLA. Zero data retention requires a separate signed agreement. The no-recharge Qwen promotion has a deadline; recharged Starter and Pro users retain tier benefits, while the model name, weights and precision may change. |
|
| Mistral Studio | Mistral's Free mode — no-card API access with an unpublished monthly allowance shared across Studio, API and the Vibe coding CLI | — |
Shared Studio, API and Vibe allowance: Monthly usage allowance: amount not published per organization …API rate limits: Per-second requests allowance: see your account for values; Per-minute tokens allowance: see your account for values per organization per model The monthly allowance is shared across Studio, API and Vibe usage in the organization; its Free-mode amount is not published. Free mode is the default for new accounts, with no credit card required. At the cap usage can stop until the next billing period unless pay-as-you-go is enabled. API throughput values are visible inside the account. API calls may be used to improve Mistral services unless the Admin panel's `Anonymous improvement data` toggle is off. |
|
| VLM Run Gateway | OpenAI-compatible gateway whose GPU-served vision and language models accept anonymous callers — no signup or key — in beta |
Anonymous lane: 10 requests/minute; 30 requests/hour; 100 requests/day per IP …GPU-served models are public and anonymous; frontier models require paid access. All three IP windows apply together and are shared by callers behind the same NAT or proxy. The gateway is beta; anonymous access is intended for evaluation. The operator is Autonomi AI Inc.; its terms render in a browser. |
||
| Cohere (trial keys) | Cohere's Command models via free trial API keys that never expire, plus a 30B/3B Apache-2.0 coding model Cohere prices at zero on every key type |
command-a-pluscommand-a-reasoningnorth-mini-codecommand-acommand-a-visioncommand-r-pluscommand-rcommand-r7b
|
Trial keys: 1,000 requests/month per key …Trial Chat models: 20 requests/minute per key per model North Mini Code on production keys: 500 requests/minute per key per model Trial keys are for evaluation, not production or commercial use; new accounts start with Trial keys. Chat limits differ from those for reranking, embeddings, tokenization and audio. North Mini Code is free on both Trial and production keys until the corresponding rate limit is reached. The trial monthly request allowance applies alongside its per-model rate caps. |
|
| SimpleLLM | OpenAI-compatible chat API from Stage One Solutions UG in Germany, with enabled zero-price aliases usable without credit |
gemma-4-e4bllama-3.1-8b
|
General: Per-minute requests allowance: see your account for values; scope not published …Free models: Hourly requests allowance: see your account for values; Daily requests allowance: see your account for values; Hourly tokens allowance: see your account for values; Daily tokens allowance: see your account for values; scope not published An account and API key are required. Request and token budgets depend on configuration; the authenticated rate-limit endpoint reports current minute, hourly and daily counters. The docs do not establish one shared scope for every key, model and upstream limit. The public catalog can list an enabled alias whose backend is disabled, so listing does not prove a successful completion. Planned or disabled rows are excluded. Numeric free limits and reset times are not publicly fixed. |
|
| SEA-LION (AI Singapore) | AI Singapore's OpenAI-compatible API for its Southeast Asian language models | — | 10 requests/minute per account The free API is intended for prototyping; production use is directed to cloud partners. The free-offer page names no model, while the model API docs publish the per-user rate. No token or credit budget or expiry is stated. | |
| OpenTyphoon (SCB 10X) | Thai-tuned open models from SCB 10X, the venture arm of Siam Commercial Bank, behind an OpenAI-compatible API whose FAQ calls it a research showcase and free to use | — |
5 requests/second; 200 requests/minute per account per model; for `typhoon-v2.5-30b-a3b-instruct` …A free research showcase, with higher limits available by email. Usage data is collected to improve the model and API; the vendor says it is not shared with third parties. Production users are directed to Together AI. A key is minted in the playground after signup; no page read mentions a card. OCR has its own lower rates, separate from the coding model. |
|
| Agnes AI | Agnes AI's own models behind an OpenAI-compatible API, with its Flash text models charged at zero today and image generation free beside them |
Allowed and effective text rates for default users: Per-minute requests allowance: see your account for values per account …The Flash models' input, cached input and output are currently zero-priced; Pro remains metered. Zero pricing is promotional: "Promotional end dates are subject to Agnes AI platform announcements and your account bill". Allowed and effective text RPM for the default user class are dynamic; check the console for current limits. Keys of the same type share a pool; enterprise and Token Plan keys use separate pools. Image rates are separate. No daily budget is published. |
||
| Zro (MoonMath) | OpenAI-compatible inference for coding agents from MoonMath, with a recurring Free plan on selected Flash models | — |
Free plan: Daily usage allowance: amount not published; scope not published …The Free plan covers the two named Flash models with a daily allowance; its amount, pooling scope, reset time and rate limits are not published. An account and API key are required. The pricing page offers Free separately from paid plans and usage packs; the general quickstart still asks for a subscription or usage pack. Inference passes through an EU gateway and runs in Europe, the US, or both as selected per key; unavailable allowed regions fail rather than falling back elsewhere. Zro says it retains no request bodies and uses no requests or logs for training without separate explicit consent; account, billing and operational metadata are retained. Inference providers follow their own operational logging terms. Optional agent tools have separate prices. |
|
| Infron |
OpenAI-compatible model gateway with a shared recurring Free request allowance, unlocked by a retained prepaid balance requires $5 top-up + 5% fee |
— |
Standard free models: 1,000 requests/day per account. 00:00 UTC …Keep at least $5 in account credit to use standard :free models. Free requests do not consume that balance. New accounts start at $0. Top-ups add a 5% platform fee plus $0.35 per transaction; payment methods include cards and Alipay. Ordinary per-account rate limits also apply. Free routes can use different providers from paid variants, share capacity and have no SLA. The separate Qwen Max Preview promotion ended on July 28, 2026. Unused purchased credit may expire after 365 days. Infron does not store prompts or responses by default; upstream retention varies and an account setting excludes providers that train on prompts. |
|
| FreeInference (Harvard SEAS) | Harvard SEAS's MadSys Lab serving open models free to every account behind both an OpenAI-shaped and an Anthropic-shaped endpoint, with a documented Claude Code setup |
Usage allowance: amount varies (period not published) per account …No card. This is an experimental research service with capacity-dependent limits. Prompts and responses may be logged for research; sanitized content, usage statistics and routing metrics may be published or open-sourced. Only models marked Free accept a Free key; Pro catalog rows need a Pro-enabled key. |
||
| Ollama Cloud | Cloud-hosted open models with a monthly starter allowance advertised in the launch announcement; current pricing instead describes pay-as-you-go usage | — |
Advertised starter usage: Monthly credits allowance: amount not published per account …The August 31 announcement still advertises monthly starter usage for a starter subset, but current pricing and download guidance describe paid cloud usage and omit that allowance. Current Free eligibility, amount and model set remain unresolved. The announcement describes signup-date monthly resets without rollover. Confirm the allowance in your account before using a cloud model; per-token prices do not establish starter eligibility. Local model execution is a separate, free use of the software. |
|
| OVHcloud AI Endpoints | EU-hosted serverless API for open models, with anonymous OpenAI-compatible access and no signup, key or card |
Anonymous lane: 2 requests/minute sharing scope disputed …The documentation attributes the anonymous cap to IP and model. Tests on 2026-09-24 instead observed callers sharing a model's anonymous capacity, so its effective sharing scope remains disputed. Anonymous requests can receive 429 before a caller has spent the documented cap. Authenticated API keys are metered per token and use the paid project/model limits; they do not create a larger free allowance. |
||
| Novita AI | Inference cloud with a zero-priced chat model through an OpenAI-compatible API, including tool calling and structured output | — |
Zero-priced model: Requests allowance: amount not published (period not published); scope not published …Apodex 1.1 Mini has zero input and output prices on its public model page, pricing table and API catalog. An account and API key are required. The general quickstart asks users to check available credit and model access; it does not state a minimum payment for this zero-priced model. No numeric Free request allowance or offer deadline is published on these pages. Other models are metered. |
|
| Sail Research | Open-weight models for long-running agents — Kimi K3, GLM-5.3, DeepSeek V4 Pro — behind OpenAI- and Anthropic-compatible APIs, with $5 of free credit every month once a payment method is on the account | — |
$5/month per account …Attaching a payment method is required for recurring free credit. Credit spends at per-token prices that depend on the requested completion window. The vendor states "No strict rate limits" and says request and response data is not used to train models without written consent. |
|
| Pollinations.AI | Legacy open text API, no signup, OpenAI-compatible (POST text.pollinations.ai/openai), on one model |
Legacy anonymous lane: Usage allowance: amount not published (period not published); scope not published …The legacy keyless host lists openai-fast, GPT-OSS 20B on OVH, in the anonymous tier. The current gen.pollinations.ai API requires a key and bills Pollen credits; without a key it returns 401. The former anonymous rate came from retired docs and is no longer a current published limit. |
||
| IBM watsonx.ai (Lite plan) | IBM's watsonx.ai Runtime on its Lite plan — 300,000 tokens a month of foundation-model inference (Granite, Llama, Mistral and other hosted models) on IBM Cloud, a plan IBM's own docs call free and never bill | — |
Lite: 300,000 tokens/month; 2 requests/second per service plan …Lite compute: 20 CUH/month per service plan The Lite service plan has limited capacity and no published expiry. IBM Cloud signup requires personal and card information to verify identity; Lite plans do not incur charges. Its covered foundation models are listed separately. The compute allowance and inference-token budget measure different kinds of usage. |
|
| Aion Labs | OpenAI-compatible text-model API with a daily credit allowance on its $0 Free tier, no card required | — |
Daily credits allowance: amount not published; 15 requests/minute; 20,000 tokens/minute; 20,000 tokens/day per account …The reset timezone is unpublished. Request and token ceilings apply separately from the credit balance; completions are billed at each model's price. An account and API key are required. The Terms call free use before purchase a courtesy that may be changed or withdrawn. |
|
| Arli AI | OpenAI-compatible inference on open models and their fine-tunes, whose Free plan tries each model five times every two days at 12K tokens of context, one request at a time | — |
5 requests/2 days per account per model …1 request at once per account Free is for testing, with delayed responses, slower service as use increases and a 12K context limit. Its catalog includes many roleplay and writing fine-tunes alongside coding-capable base models. Continuing beyond the trial allowance requires a paid plan. |
|
| Bytez | Serverless API over open models, whose Free plan grants $1 of credit every four weeks for open models of up to 7B parameters, one request at a time, with no billing | — |
Open models up to 7B: $1/4 weeks; 1 request at once per account …Credits expire four weeks after grant. The free balance supports open models up to 7B parameters. Closed-model docs conflict between provider prices plus a platform fee and BYOK billed directly by the provider, so the free dollar does not establish bundled closed-model access. Larger open models require paid credit and a paid plan. |
|
| Mixlayer | Serverless open models priced per token, one of them at $0 and callable without prepaid credit |
Usage allowance: see your account for values (period not published) per organization per model …Free models do not need prepaid credit; paid models return 402 when the balance is empty. The zero-priced Qwen route is separate from metered models. Rate limits are set per organization and model, with actual values unpublished. The operator is Mixlayer Labs Inc. |
||
| ModelScope API-Inference (Alibaba) | Alibaba model community with a free daily API inference allowance measured in credits, after Alibaba Cloud real-name verification | — |
Magicube daily login: 200 credits/day per account (Valid for the day) …Daily bound-account bonus: 50 credits/day per account (Alibaba Cloud account bound; valid for the day) Additional community rewards: Credits allowance: amount varies (period not published) per account API inference requires a bound Alibaba Cloud account with real-name verification; spending Magicube also requires a verified personal email. Calls cost 0.5, 1 or 2 credits by model class, so credits are not a fixed request count. Community rewards are separate and variable. Concurrency follows platform load. This is non-commercial, unsuitable for production requiring high concurrency or SLA; older models can be replaced. |
|
| SenseNova (SenseTime 商汤) | SenseTime's SenseNova API, with a $0 Free plan during public beta; registration and identity verification required |
sensenova-6.8-flash-lite
|
Public beta: Usage allowance: amount not published (period not published) per account …Free during the public beta; paid tiers are planned. The Free card covers SenseNova 6.8 Flash Lite and SenseNova U1 Fast. Current numeric usage allowances and reset times are unpublished. The API guide calls the documented model an "early preview release" and requires "registration plus identity verification". Up to 20 API keys are supported. Signup requires a phone number; acceptance of non-mainland numbers is unverified. |
|
| Tencent Cloud TokenHub | Tencent Cloud's model platform, with a one-time grant of a million tokens on each of its language models, valid a year and no card, on an account that has passed Tencent Cloud real-name verification | — |
Claimed promotional package: 1,000,000 tokens once per account per model (Valid for 1 year) …The claiming promotion runs through December 31, 2026. Grants are selected per model or claimed on its first call; one note instead says models share the account's quota, so confirm the console's scope. Exhaustion stops usage unless post-payment is enabled. Real-name verification accepts residents of mainland China, Hong Kong, Macau and Taiwan; other identity documents are directed to the international service. |
|
| SiliconFlow (China) | Zero-priced small chat models on SiliconFlow China, after real-name verification with accepted mainland, Hong Kong, Macau or Taiwan documents |
Verified free models: Usage allowance: see your account for values (period not published) per account per model …Free models are zero-priced but require real-name verification. Online verification uses an Alipay face scan and accepted mainland, Hong Kong, Macau, Taiwan or Chinese permanent-residence documents; other applicants must contact staff. Sign in by SMS or email. Per-model fixed rate values are shown in the client-rendered model square, rather than public docs. |
🎁 Trials (no card when possible)
19 live · not one of them asks for a card · sorted by how much work you can get done for free
| Tool | What you get | Free models | Limits, in the vendor's words | Verified |
|---|---|---|---|---|
| Gemini Enterprise Agent Platform express mode (formerly Vertex AI) | Google Cloud's express mode: an API key and 90 days of Gemini models within the free tier's quotas, with no billing information, for a new Google Cloud user on a @gmail.com account |
Preview models: Per-minute requests allowance: amount varies per project per model …Stable Gemini models: 10 requests/minute per project per model (Free express trial lasts up to 90 days); for `gemini-2.5-pro`, `gemini-2.5-flash` Preview Gemini models: Per-minute requests allowance: amount varies per project per model (Free express trial lasts up to 90 days); for `gemini-3.1-pro-preview`, `gemini-3-flash-preview` The free express trial is for new Google Cloud users and requires no billing information. Stable and Preview model rate policies differ; enabling billing is necessary after the trial. Existing Google Cloud accounts do not get this offer. The separate $300 Cloud Free Trial needs a payment method. Express is Preview and may not process personal data. |
||
| Regolo AI | EU (Italian) zero-retention inference; a month of full model access on a daily token allowance, no card |
30-day trial: 1,000,000 tokens/day per account …"Start your 30-day free trial ... No credit card required, no commitment": the trial card names "1 month duration" ("Full access for 30 days, then choose a plan"), and "Stricter rate limits" with "Fair usage throttling applies", against "All Core Models", which on the same page is every chat model in the library table (each marked Included under Core). Nothing survives the 30 days — the page names no grant after it, only paid plans — and the daily figure is the only number the trial publishes. One model is priced at €0.00 in and out outside any trial, brick-v1-beta, and it is not one to code with: its own page calls it "a lightweight prompt-complexity classifier designed for LLM routing pipelines", a Qwen3.5-0.8B LoRA that labels a prompt easy, medium or hard for Regolo's Brick router (read 2026-09-21) |
||
| Upstage (Solar API) | Upstage Solar LLM API; $10 free credit on signup, no card | — | Signup credit: $10 once per account See the console for credit validity. After the signup credit is spent, continue with prepaid credit or pay-as-you-go billing. | |
| Zed | Open-source code editor with a hosted AI agent: a 14-day Pro trial with $5 of model usage and unlimited edit predictions, no card, plus a separate year of free Student access for verified university students | — |
Shared Zed and Delta trial: $5 once per account (Valid for 14 days) …14-day trial: Unmetered edit predictions per account Verified student plan: $10/month; Unmetered edit predictions per account (Free for one year to verified university students) No card for the trial. GPT-6 Luna is the only hosted model during the trial, and the shared Zed/Delta trial ends when the credit or time runs out, whichever comes first. It then converts to Free, which uses your own API keys or external agents. The docs say "Zed's hosted models and edit predictions require a Pro subscription." Verified university students receive a separate year of free Student access. Token credit is a spending wallet at hosted-model prices, not a list of individually zero-priced models. |
|
| AtomCode | Open-source terminal coding agent from AtomGit, the code host run by CSDN and the Open Atom Foundation, whose sign-in claims a free 30-day coding plan, its quota counted in five-hour windows and its claims capped at a thousand a day |
Claimed Lite plan: Rolling 5-hour usage allowance: amount not published per account (Valid for 30 days) …Claimed Lite lasts 30 days; claims have a platform-wide pool of 1,000 slots per day, not a per-user inference cap. GLM-5.2 needs a separate Pro claim: "Pro Trial 7 days Free claim, 100 slots/day at 10:00" or "Pro 30 days Free claim, requires a merged PR in the activity repository". All plan usage is measured in rolling five-hour windows without published numerical budgets. CodingPlan works inside AtomCode via AtomGit sign-in; other providers use your own key. |
||
| ZCode (Z.ai) | Z.ai's desktop coding agent, free as an app, with five days of its GLM models for a new user and no card |
Global trial: GLM-5.3: 3,000,000 tokens/day per account per model (First 5 days only) …Global trial: GLM-5-turbo: 2,000,000 tokens/day per account per model (First 5 days only) BigModel trial: GLM-5.3: 3,000,000 tokens/day per account per model (First 5 days only) BigModel trial: GLM-5.3-Flash: 5,000,000 tokens/day per account per model (First 5 days only) Global and mainland BigModel accounts have different per-model daily trial grants. The global trial covers GLM-5.3 and GLM-5-Turbo; BigModel covers GLM-5.3 and GLM-5.3-Flash in separate buckets. Both end after the first five days and do not create ongoing daily access. Afterwards the app needs a GLM Coding Plan or your own API key; the application itself remains free. |
||
| Qodo | Agentic PR code review plus Git and IDE integrations on a 14-day Pro Team trial with unlimited reviews and credits and no card | — |
14-day trial: Unmetered credits per team (Free for 14 days) …Open Source program: Usage allowance: amount not published (period not published) per project (Qualified projects; application required) No card for the trial. There is no permanent general-purpose free tier; continued reviewing needs a paid plan, except qualified open-source projects that apply to the standing free program. The trial's models are unnamed. The terms permit customer data to train and improve the platform and models, while the pricing FAQ says Qodo does not train on code; those statements conflict. |
|
| The Grid | OpenAI- and Anthropic-compatible inference market that sells quality tiers rather than model names — Agent Max was served by Claude Opus 5 in the 30 days to 2026-09-03 — with a $25 signup credit, for a limited time | — |
Promotional signup credit: $25 once per account …Before first deposit: 100 requests/day; 3 requests at once per account The signup grant is a limited-time offer with no payment step. A first deposit lifts the daily free cap; concurrency counts deposits, excluding signup bonus. Free-tier prompts and completions are stored and stay stored after upgrade; zero retention is paid-only. Model names buy specifications, and the underlying model can change between calls. Historical catalog deliveries do not guarantee the next call's model. |
|
| Dahl Inference | An OpenAI-compatible gateway to open models served by the Gonka decentralized GPU network — GLM-5.3-Flash, DeepSeek V4 Flash and MiniMax M2.7 — whose sign-up asks for a username and nothing else and puts 100 million tokens in the account | — |
100,000,000 tokens once per account …Signup tokens enter the account pool; a new key starts empty and returns 402 until tokens are allocated at /account. Extra keys do not grant extra tokens. Save the fingerprint shown once: it is the password, with no email recovery. Terms prohibit automated account creation to harvest grants and permit ending discretionary free access. Further top-ups are crypto. Requests go to independent decentralized operators under FROMZERO OÜ's terms. |
|
| Sarvam AI | India's Sarvam AI grants verified users ₹100 in credits that never expire, spendable across its APIs, including Sarvam-105B chat through an OpenAI-compatible endpoint | — |
₹100 once per account …Pay-as-you-go Sarvam-105B: 60 requests/minute per account (Shared by all API keys under the account); for `sarvam-105b` Signup credit requires a verified user, applies across APIs and never expires; its value is in INR, with no guaranteed token conversion. Rate limits apply per account across its keys, with separate pools by API. The listed Pay-as-you-go rate covers Sarvam-105B. The v2 chat and Responses APIs are generally available for the documented open models; the pricing page still labels some other models beta. No page read mentions a card. |
|
| Inception Labs (Mercury) | A one-time signup token balance on Mercury diffusion models, including the current Mercury 3 chat model and earlier chat and code-edit models | — |
100,000,000 tokens once; 1,000 requests/minute; 1,000,000 input tokens/minute; 100,000 output tokens/minute per account …No payment details are required. The signup token balance is shared across all models, does not refill and is spent alongside the Free tier's per-minute traffic limits. After it, usage is pay-as-you-go at the model's current token prices; a launch discount is not a separate free model. |
|
| HPC-AI Model APIs | OpenAI-compatible inference with $2 of signup credit, or $4 in total with the vendor invite code, before any deposit | — |
Standard signup grant: $2 once per account …Signup grant with invite code: $4 once per account (Use invite code HPCAI-MAPI) L0 before first deposit: 5 requests/minute per account per model; for `zai-org/glm-5.3-flash`, `moonshotai/kimi-k2.7-code`, `zai-org/glm-5.3`, `deepseek/deepseek-v4-flash`, `anthropic/claude-fable-5`, `anthropic/claude-opus-4.7`, `anthropic/claude-opus-4.8`, `anthropic/claude-opus-5`, `moonshotai/kimi-k3`, `openai/gpt-5.5`, `openai/gpt-oss-120b`, `qwen/qwen3.8-2.4t-a95b`, `qwen/qwen3.8-max`, `zai-org/glm-5.2`, `anthropic/claude-opus-4.6`, `xiaomi/mimo-v2.5-pro`, `xiaomi/mimo-v2.5`, `moonshotai/kimi-k2.6`, `nvidia/nemotron-3-ultra-550b-a55b`, `minimax/minimax-m3`, `qwen/qwen-3.5-397b-a17b`, `qwen/qwen-3.5-35b-a3b`, `qwen/qwen-3.5-27b` Standard L0 models: 2,000,000 tokens/minute per account per model; for `zai-org/glm-5.3-flash`, `moonshotai/kimi-k2.7-code`, `zai-org/glm-5.3`, `deepseek/deepseek-v4-flash`, `anthropic/claude-fable-5`, `anthropic/claude-opus-4.7`, `anthropic/claude-opus-4.8`, `anthropic/claude-opus-5`, `moonshotai/kimi-k3`, `openai/gpt-5.5`, `openai/gpt-oss-120b`, `qwen/qwen3.8-2.4t-a95b`, `qwen/qwen3.8-max`, `zai-org/glm-5.2`, `xiaomi/mimo-v2.5-pro`, `xiaomi/mimo-v2.5`, `moonshotai/kimi-k2.6`, `nvidia/nemotron-3-ultra-550b-a55b`, `minimax/minimax-m3`, `qwen/qwen-3.5-397b-a17b`, `qwen/qwen-3.5-35b-a3b`, `qwen/qwen-3.5-27b` L0: Claude Opus 4.6: 10,000,000 tokens/minute per account per model; for `anthropic/claude-opus-4.6` The invite-code grant is the total signup balance, not an extra amount added to the standard grant; supplies are limited. Credit spends at model list prices. L0 applies before the first deposit; spending alone does not promote an account to L1. Claude Opus 4.6 has a higher token rate than the other positive-RPM L0 models; DeepSeek V4 Pro has zero L0 RPM. No card requirement or credit expiry is published. |
|
| Fireworks AI | Serverless inference on open-weight models with a $1 trial credit documented in the billing FAQ; current pricing no longer advertises a signup grant | — |
Documented credit; signup eligibility unverified: $1 (period not published) (published sources disagree) per account …10 requests/minute per account The billing FAQ describes a credit and account suspension after exhaustion without a payment method. Current pricing instead emphasizes prepaid credit and does not promise a signup grant; new-account eligibility remains unverified. No expiry or recurrence is stated. Credit spends at list token prices. The low account rate applies when no payment method or no credit is present. |
|
| RouterPlex | A one-time $1 of free credit on a prepaid reseller that bills its catalog at vendor list prices with 0% markup, no card — one key on both wires, OpenAI-compatible Chat Completions and an Anthropic Messages base Claude Code takes as it is | — |
Key-creation grant: $1 once per account …Trial traffic: 10 requests/minute; 250,000 tokens/minute; 4 requests at once per account The key-creation grant is for an account with no earlier paid top-up, existing credit or setup grant; signup has no payment step and the grant does not expire. It spends at catalog prices with no token markup. Before a first paid top-up, only an unpublished subset of models may accept promotional credit, at reduced rates. Terms do not guarantee a grant and allow one account per person. Continuing requires at least $5 by card or $15 by crypto. No legal entity is named; the catalog needs a key. |
|
| Wallaby | OpenAI-compatible Kimi K3 API from an Australian open-weight inference gateway, with $0.50 of signup credit and no card or initial payment | — |
Signup trial credit: $0.5 once per account (No card or top-up required) …Trial rate limit: Requests allowance: amount not published (period not published) per account One trial per person or organisation, with a verified email and human check. The trial credit does not expire; requests stop when it is exhausted. Optional paid top-ups start at $20. Trial access may have lower rate limits or model restrictions; the vendor publishes no numeric trial rate. Wallaby says it does not log or train on prompts, but its privacy policy does not establish the upstream providers' retention or training rules. Users must be at least 18. |
|
| abliteration.ai | OpenAI- and Anthropic-compatible API for uncensored reasoning models, with a one-credit free preview and no card | — |
1 credit once per organization …Sign in to use the preview without a card. A preview credit's fiat value is unpublished even though the organization balance endpoint uses USD; no conversion is inferred. Further usage needs prepaid credit or a paid plan and is billed per token. The models stream, call tools and think before answering by default. |
|
| Impossibl | Prepaid gateway at provider list prices, Claude, GPT, Gemini, DeepSeek and GLM among its models, whose keyless sign-up funds an account with $0.05 and adds $1 once a person claims it by email — no card | — |
Signup credit: $0.05 (period not published) per account …Email-claim grant: $1 once per account An account can be created by a keyless POST; email sign-in claims the additional grant once per person. Some models require a completed credit purchase and return `403 billing_required` without it; adding a card or holding promotional credit does not meet that condition. The eligible subset is unpublished. Usage bills at provider list prices without token markup; top-ups start at $5 plus a 5% fee. |
|
| Standard Compute | A one-time $0.25 of smart-routed compute on a flat-rate agent gateway, no card — a real key on both wires, OpenAI-compatible Chat Completions and an Anthropic Messages base Claude Code takes as it is | — |
Platform compute: $0.25 once per account …A finite platform-compute trial, not a recurring budget or fixed token count. No card is required, but eligibility and availability are discretionary; contact support if activation offers no grant. No expiry or per-token conversion is published. The vendor recommends a small connection test and warns that a longer coding comparison may need paid allowance. |
|
| CodeAnt AI | Installed CLI for AI pull-request review and suggested fixes on existing code, with a no-card introductory trial | — |
One-time usage allowance: amount not published per account …The "14-day free trial" requires no card. The CLI authenticates with your code-host account. Paid plans are offered after the trial. The Trust Center says LLM requests are "not logged or retained by CodeAnt AI or the model provider"; personal information can be retained while your account is active. |
🧭 Aggregators (one key, many providers)
16 live · 14 of them ask for no card · sorted by how much work you can get done for free
| Tool | What you get | Free models | Limits, in the vendor's words | Verified |
|---|---|---|---|---|
| OpenRouter (free models) | One API key for a rotating set of :free model variants, open-weight and stealth models among them |
50 requests/day; 20 requests/minute per account …1,000 requests/day per account (after 10 credits purchased all-time) OpenRouter's FAQ says its free models "have low rate limits" and "are usually not suitable for production use", and its limits page warns that a negative credit balance can produce errors "including for free models" and that a 429 can come from the upstream provider rather than the platform (read 2026-09-27). The docs define the daily counter as "Free-model requests recorded so far in the current UTC day". The higher daily ceiling is granted "starting one credit below the table’s threshold" to absorb rounding and top-up fees. |
||
| Requesty | OpenAI-compatible router with free models beside a metered catalog, routing, caching and fallbacks | 200 requests/day per account No credit card for the Free plan, which serves free models only. Routing, caching, fallbacks, spend tracking and EU data residency are included. Past the free allowance the same key moves to pay-as-you-go. | ||
| AIHubMix (free models) | OpenAI-compatible model gateway with zero-priced routes and an Anthropic-format endpoint for coding clients |
100/day coding routes: 5 requests/minute; 100 requests/day; 1,000,000 tokens/day per account per model; for `xiaomi-mimo-v2.6-pro-free` …100/day coding routes: 5 requests/minute; 100 requests/day; 1,000,000 tokens/day per account per model; for `xiaomi-mimo-v2.6-pro-free`, `coding-glm-5.3-flash-free`, `coding-glm-5.3-free`, `xiaomi-mimo-v2.6-flash-free`, `coding-glm-5.2-free`, `coding-kimi-k3-free`, `coding-glm-5.1-free`, `coding-minimax-m2.7-free`, `coding-glm-5-free`, `coding-glm-5-turbo-free`, `coding-minimax-m2.5-free` 500/day coding routes: 5 requests/minute; 500 requests/day; 1,000,000 tokens/day per account per model; for `coding-minimax-m3-free`, `glm-4.7-flash-free`, `coding-glm-4.7-free`, `k2.6-code-preview-free`, `coding-minimax-m2.1-free`, `kimi-for-coding-free`, `coding-glm-4.6-free`, `coding-minimax-m2-free` Other free routes: Usage allowance: amount not published (period not published); scope not published; for `agents-a1-free`, `intern-s2-free`, `dots-3-note-preview-free`, `lfm-2.5-2.6b-free`, `nemotron-3.5-lightning-free`, `nemotron-3-super-120b-a12b-free`, `nemotron-3-nano-omni-30b-a3b-reasoning-free`, `nemotron-3-ultra-550b-a55b-free`, `north-mini-code-free`, `mimo-v2-flash-free` Gemma 4 free routes: Usage allowance: amount not published (period not published); scope not published; for `gemma-4-26b-a4b-it-free`, `gemma-4-31b-it-free` Free IDs end in -free; paid twins are metered at list prices. The daily request and token caps are independent per account and model, rather than a pool to split across models. Published caps apply to their named model IDs; the remaining routes have no numerical budget published. Request limits and campaign durations apply to their named IDs. nemotron-3.5-content-safety-free is a classifier; the lfm-2.5-2.6b-free developer advises against agentic coding use. |
||
| Routeway | OpenAI-compatible gateway with a rotating :free chat-model lane beside metered models |
Rate-limits documentation: 5 requests/minute (published sources disagree) per account …FAQ: 20 requests/minute (published sources disagree) per account Free models: 200 requests/day per account The FAQ and rate-limit docs disagree on the per-minute cap; both publish the same daily cap. Free IDs end in :free, rotate and may be removed at any time. At a cap they return 429; paid twins require a positive balance. The FAQ describes signup and key creation without payment, but the older terms list a payment method. No legal entity is named; support is email and Discord. |
||
| LLMTR | Turkish OpenAI-compatible gateway with free models on a zero balance and dated previews; some previews require a previous top-up |
nemotron-3-ultraagnes-3.0-flashagnes-2.5-flashqwen3.8-27bjais-3-mova-36b-a4buntil 2026-11-05k2-horizon-375b-a23buntil 2026-11-05minimax-m3.1-flashuntil 2026-11-02nemotron-3.5-super-vl-120b-a12b
|
Nemotron Ultra and Qwen3.8 27B: Daily usage allowance: amount not published; scope not published; for `nvidia/nemotron-3-ultra-550b-a55b` …Nemotron Ultra and Qwen3.8 27B: Daily usage allowance: amount not published; scope not published; for `nvidia/nemotron-3-ultra-550b-a55b`, `qwen/qwen3.8-27b-free` Nemotron3.5 Super VL: Daily usage allowance: amount not published; scope not published; for `nvidia/nemotron-3.5-super-vl-free` Jais 3 MoVA: 200,000 tokens/day per account per model (Input and output count together); for `ifm/jais-3-mova-36b-a4b`. 00:00'da (TSİ) K2 Horizon: 200,000 tokens/day per account per model (Input and output count together; reasoning counts as output); for `ifm/k2-horizon-375b-a23b`. 00:00'da (TSİ) MiniMax M3.1 Flash: 5,000,000 tokens/day per account per model; for `minimax/minimax-m3.1-flash-free`. 00:00’da (TSİ) Free chat rows can be called on a new account with zero balance before any top-up. Paid use is prepaid; an 8% margin applies once at top-up, not to model prices. Prompt and response bodies are not permanently written to the usage and billing database. Heabsy requires provider-specific data-transfer consent: "Heabsy models cannot be called until your account has granted the data-transfer consent for Heabsy". Its stream is buffered until completion, and reasoning_effort is unsupported. For free Nemotron3.5 Super VL, inputs "may be retained by the provider and used to improve the model". Shanghai AI Lab prompts and completions may be logged by the provider and used to improve models. |
|
| Vercel AI Gateway | OpenAI-compatible model gateway with $5 of monthly credit after adding a payment method, plus separate zero-priced model routes |
Gateway credit: $5/month per team …Free credit rate limits: Requests allowance: amount not published (period not published); scope not published (Limits apply per model.) Beam launch offer: Usage allowance: amount not published (period not published); scope not published (The free route stops at the end of the launch offer; the end date and numerical allowance are unpublished.); for `reflection/beam-free` A team must add a valid payment method before using free credit; otherwise requests return `403 customer_verification_required`. Credit spends at provider list rates and excludes BYOK. Buying credit ends the monthly free credit. Zero-priced model routes remain a separate free mode and do not spend that balance. Use their exact free IDs: a similarly named route without its free suffix can be metered. |
||
| BazaarLink | OpenAI-compatible gateway with a shared free allowance on selected models and an auto:free router |
Without credit: 10 requests/rolling minute; 60 weighted units/day per account …With credit: 20 requests/rolling minute; 120 weighted units/day per account Free models: 2 requests at once per account Free models: 20 requests/minute shared across the endpoint Additional IP guard: Requests allowance: amount not published (period not published) per IP The free models share each account's daily weighted allowance, reset at 00:00 UTC. Longer inputs can consume more units. Account concurrency and the site-wide rate also apply. Paid fallback happens only when enabled and the account has sufficient credit. Free IDs and their metered twins are separate catalog routes. |
||
| Inferbase | OpenAI-compatible managed inference on a permanent $0 Free plan, with a shared daily usage budget and no card required | — |
All successful requests: 5,000 requests/month per organization …Managed inference: 100 requests/day; $0.5/day; 10 requests/minute per organization 60 requests/minute per key Managed chat, embeddings and rerank share the daily allowance. BYOK requests also count toward the monthly request ceiling, but their tokens are billed by the connected provider. Managed requests allow at most 8,192 input and 1,024 output tokens; these are request budgets, not full model context windows. Monthly requests reset on the 1st at 00:00 UTC. Exhausted allowances return 402. Managed image, audio and video generation require Pro. |
|
| Nous Portal (Hermes Agent) | Nous Research's inference portal with a $0 Free plan for :free models, accessible through an OpenAI-compatible API and Hermes Agent |
Free models: Usage allowance: amount not published (period not published) per account …Choose the Free plan and use exact :free IDs; they are zero-priced, while metered routes are outside it. Use a portal key with ordinary API clients; the catalog itself is public. The Free plan publishes no card requirement or numerical rate cap. |
||
| Token Harbor | OpenAI- and Anthropic-compatible gateway with a standing $0 plan: a rotating lineup of :free ids on a value-based allowance per rolling 7-day period, no card |
Free routes: Rolling 7-day list-price value allowance: amount not published per account …Free accounts: 60 requests/minute; 1,800 requests/hour per account Free accounts: 100 requests/minute; 3,000 requests/hour per IP Enable explicit :free routes in the dashboard before calling them; the rotating model set is listed under Free. The dashboard shows usage as a percentage. Exhausted free routes stop instead of charging the wallet. Request-rate caps apply to Free accounts across models and API keys. "There is no limit on concurrent requests, on any plan." "Free accounts: up to 4 MB (4,000,000 bytes)." Oversized requests return `413 request_too_large`. Permanent free routes require data sharing: staff and third-party AI tools may analyze retained prompts and responses for diagnostics and model or product improvement. Token Harbor says it does not train models itself; upstream providers may train under their own terms. Limited-time campaigns may expressly use the paid-route data policy instead. The operator is Token Harbor PTE. LTD., Singapore. |
||
| TokenRouter (PaleBlueDot) | OpenAI-compatible gateway with a zero-priced free-model route in its default group beside metered models |
Free route: Usage allowance: amount not published (period not published) per account …The free route belongs to the default group. Other catalog routes are metered, apart from an unmarked zero-priced stealth route. No numerical free request or token budget is published. |
||
| Opper | EU-hosted model gateway advertising a free route before adding a card; current documentation conflicts on its availability |
Advertised route; free access unconfirmed: Usage allowance: published sources disagree (period not published); scope not published …The signup guide advertises a free US-hosted Gemma route before adding a card, but the current directory has no free routes and its Gemma page lists paid alternatives. The API catalog retains the ID without a price or availability claim. Free access remains unconfirmed. Paid credit purchases add a 3% fee. Opper Technology AB operates in Sweden on AWS Stockholm. |
||
| MegaNova | OpenAI-compatible gateway with daily free model quotas: no-card Tier 1, plus additional free access after a one-time deposit |
mistral-small-3.2glm-4.7-flashrequires $1 top-up
|
Tier 1: Manta Mini 1.0: 50 requests/day per account per model; for `meganova-ai/manta-mini-1.0` …Tier 1: Manta Flash 1.0: 50 requests/day per account per model; for `meganova-ai/manta-flash-1.0` Tier 1: Mistral-Small-3.2-24B-Instruct-2506: 50 requests/day per account per model; for `mistralai/Mistral-Small-3.2-24B-Instruct-2506` Tier 2: Manta Mini 1.0: 500 requests/day per account per model (After a $1 deposit); for `meganova-ai/manta-mini-1.0` Tier 2: Manta Flash 1.0: 500 requests/day per account per model (After a $1 deposit); for `meganova-ai/manta-flash-1.0` Tier 2: Mistral-Small-3.2-24B-Instruct-2506: 300 requests/day per account per model (After a $1 deposit); for `mistralai/Mistral-Small-3.2-24B-Instruct-2506` Tier 2: Manta Pro 1.0: 50 requests/day per account per model (After a $1 deposit); for `meganova-ai/manta-pro-1.0` Tier 2: GLM-4.7-Flash: 50 requests/day per account per model (After a $1 deposit); for `zai-org/GLM-4.7-Flash` Tier 1: 60 requests/minute; 200,000 tokens/minute per account Tier 1 registration needs no card. A $1 deposit unlocks Tier 2's separate per-model free allowances. Daily quotas reset at 00:00 UTC; they are per model, not one total budget to spend on any model. Free modules are for evaluation and interactive use, not production or unattended batches. The operator is Nebula Nova Inc. Audio and embedding quotas do not establish coding-model allowances. |
|
| Moark (Gitee AI) | Gitee's model platform, formerly Gitee AI — 200+ open models behind OpenAI- and Anthropic-compatible APIs — whose free experience token gives every user 100 calls a day across its featured models, with nothing bought | — |
100 requests/day per account …Choose the free experience token; it can be used without buying a resource package. At the daily cap it returns `400 已达到最大当日免费 API 使用次数，请购买资源后继续使用 API`. Featured models are listed in the client-rendered model square. Its separately marked free models require a purchased resource package. Sign in through Gitee; acceptance of non-mainland phone numbers is unverified. |
|
| NovAI | Hong Kong OpenAI-compatible gateway with a recurring free allowance on selected text and vision models, separate from its signup credit | — |
Shared free-model calls: 10 requests/day per account (before the first completed top-up). 00:00 UTC+8 …5 requests/minute per account (before the first completed top-up) Separate signup credit: $2 once per account (valid for 14 days) Free chat calls: Unmetered requests per account (after any completed top-up; upstream provider limits still apply) The daily allowance is shared across free chat, image and video models; media keep separate caps after a top-up. Calls at a cap return 429 until reset. An email account and API key are required. Signup accepts selected email providers and is screened for abuse. Paid models and top-ups are optional. The gateway says it does not store or train on prompts; upstream providers apply their own policies. Users must be at least 18. |
|
| Experiential Labs |
Open-source AI gateway backed by Y Combinator; its pricing page advertises a recurring Free core-model pool, separate from paid wallet credits requires $1 card verification |
— |
Free core models: 10 requests/minute; 250,000 tokens/minute; 1,000,000 tokens/day per service plan. reset at 00:00 UTC …The pricing page advertises Luna and DeepSeek chat in the Free core pool, but their current model catalog applies zero-price plan discounts only on paid plans. This eligibility conflict remains unresolved. The earlier Astra and Fable promotional free tiers have ended. Free has no general monthly credits. A temporary $1 card verification is refunded rather than credited to the wallet. Clef and Luna Decisions use decision APIs rather than chat. Free prompt and response storage cannot be disabled without a paid plan. |
🧠 Looking for one model in particular?
126 model families, and everyone who serves each one free. A model gets a page of its own once two rows or more serve it free, or it measures notable, strong or frontier, or it has a dated free promotion.
Show the model index
🔧 Plug it into your agent
Base URLs and keys for opencode, aider, Cline and OpenAI SDKs.
Anthropic-format URLs use the same key with Claude Code's ANTHROPIC_BASE_URL.
For Codex CLI, use a listed direct profile or the LiteLLM profile below.
Show the connection table for all 62 APIs
| Provider | Base URL | Key | Notes |
|---|---|---|---|
| Google AI Studio (Gemini API) |
https://generativelanguage.googleapis.com/v1beta/openai/
|
GOOGLE_AI_STUDIO_API_KEY
|
Pass the key as Bearer. The 2.5 IDs require projects that previously used those models; new projects should choose the current Flash or Gemma IDs. |
| OpenRouter (free models) |
https://openrouter.ai/api/v1
Anthropic format:
https://openrouter.ai/apiCodex CLI:
codex -p openrouter-free with its profile |
OPENROUTER_API_KEY
|
Use IDs with the :free suffix; their input and output prices are zero. openrouter/free chooses a free model automatically. LiquidAI advises against …Use IDs with the :free suffix; their input and output prices are zero. openrouter/free chooses a free model automatically. LiquidAI advises against agentic coding with lfm-2.5-2.6b. For Claude Code, set ANTHROPIC_BASE_URL=https://openrouter.ai/api, leave ANTHROPIC_API_KEY empty and use a :free ID as ANTHROPIC_MODEL. |
| Groq |
https://api.groq.com/openai/v1
|
GROQ_API_KEY
|
the chat models Groq's own Free Plan Limits table names; its safeguard and prompt-guard classifiers, compound systems, whisper and orpheus rows are left out |
| Kilo Code |
https://api.kilo.ai/api/gateway
Codex CLI:
codex -p kilo-code with its profile |
none — the lane is anonymous | Use free IDs without a key; metered IDs require sign-in. kilo-auto/free and openrouter/free choose a free model automatically. LiquidAI advises against agentic coding with lfm-2.5-2.6b. |
| Requesty |
https://router.requesty.ai/v1
Anthropic format:
https://router.requesty.ai |
REQUESTY_API_KEY
|
Free-plan IDs are priced zero and have no expired retirement date; they carry no :free suffix. NVIDIA rows disclose training use and 30-day …Free-plan IDs are priced zero and have no expired retirement date; they carry no :free suffix. NVIDIA rows disclose training use and 30-day retention. For Claude Code, set ANTHROPIC_BASE_URL=https://router.requesty.ai, or https://router.eu.requesty.ai for EU residency. |
| NVIDIA NIM (build.nvidia.com) |
https://integrate.api.nvidia.com/v1
|
NVIDIA_NIM_API_KEY
|
Use available chat endpoints marked Free Endpoint. /v1/models also lists older IDs without a free mark, so catalog membership alone does not establish free access. |
| AIHubMix (free models) |
https://aihubmix.com/v1
Anthropic format:
https://aihubmix.comCodex CLI:
codex -p aihubmix with its profile |
AIHUBMIX_API_KEY
|
The listed -free IDs are zero-priced. intern-s2-free serves Intern-S2-Preview 35B; k2.6-code-preview-free describes kimi-for-coding. For Claude Code, use https://aihubmix.com as ANTHROPIC_BASE_URL. |
| Z.ai (Zhipu GLM) |
https://api.z.ai/api/paas/v4
Anthropic format:
https://api.z.ai/api/anthropic |
ZAI_GLM_API_KEY
|
the ids are the chat completion reference's own, where glm-4.7-flash and glm-4.5-flash sit in its model enum; Coding-Plan keys use …the ids are the chat completion reference's own, where glm-4.7-flash and glm-4.5-flash sit in its model enum; Coding-Plan keys use https://api.z.ai/api/coding/paas/v4 instead. The Claude Code guide sets ANTHROPIC_BASE_URL to https://api.z.ai/api/anthropic with "your_zai_api_key" as the token and names Coding-Plan ids (glm-5.3, glm-5.3-flash); whether the free Flash ids answer on that route is not stated, and the route answers 401 keyless (2026-09-05) |
| Cloudflare Workers AI |
https://api.cloudflare.com/client/v4/accounts/{account_id}/ai/v1
|
CLOUDFLARE_WORKERS_AI_API_KEY
|
substitute {account_id} with your Cloudflare account ID |
| Routeway |
https://api.routeway.ai/v1
Codex CLI:
codex -p routeway with its profile |
ROUTEWAY_API_KEY
|
Only IDs ending :free are free; the same catalog meters Claude, GPT and other paid models. Some zero-priced community fine-tunes are intended for expressive writing and roleplay. |
| LLM7.io |
https://api.llm7.io/v1
|
LLM7_API_KEY
|
Use a free dashboard token for the documented allowance. The listed IDs are marked turbo with usage_based_only false in the catalog; their authenticated availability has not been independently confirmed. |
| LLMTR |
https://llmtr.com/v1
|
LLMTR_API_KEY
|
New accounts have a lower free allowance for 24 hours. Heabsy needs Settings → Data-transfer consents, buffers streams and rejects reasoning_effort …New accounts have a lower free allowance for 24 hours. Heabsy needs Settings → Data-transfer consents, buffers streams and rejects reasoning_effort. Nemotron3.5 VL: omit reasoning_effort; reasoning: false disables thinking. Jais 3: low/high (default high), tool_choice: auto. K2 Horizon: medium/high; thinking is mandatory. Shanghai AI Lab: use reasoning: false; omit reasoning_effort. Intern-S2 (397B) requires tool_choice: auto and has no response_format. Include reasoning in max_tokens. |
| Vercel AI Gateway |
https://ai-gateway.vercel.sh/v1
Anthropic format:
https://ai-gateway.vercel.shCodex CLI:
codex -p vercel-ai-gateway with its profile |
VERCEL_AI_GATEWAY_API_KEY
|
Use the IDs listed here; a zero catalog price can depend on the provider and change after a promotion. For Beam, use reflection/beam-free: it stops …Use the IDs listed here; a zero catalog price can depend on the provider and change after a promotion. For Beam, use reflection/beam-free: it stops when free access ends. The regular reflection/beam route can charge at provider rates. Reflection's terms and privacy policy apply. For Claude Code, set ANTHROPIC_BASE_URL=https://ai-gateway.vercel.sh and leave ANTHROPIC_API_KEY empty. |
| Hetzner Inference API |
https://inference.hetzner.com/api/v1
|
HETZNER_INFERENCE_API_KEY
|
both ids are 262K-context and take text and images; /v1/models, /v1/completions and /v1/chat/completions are the whole surface. The key is minted in …both ids are 262K-context and take text and images; /v1/models, /v1/completions and /v1/chat/completions are the whole surface. The key is minted in the Inference tab of experiments.hetzner.com, which is a client-rendered page behind a Hetzner login — the readable copy of the terms is the docs page this row probes |
| Alibaba Cloud Model Studio (DashScope, international) |
https://dashscope-intl.aliyuncs.com/compatible-mode/v1
|
ALIBABA_MODEL_STUDIO_API_KEY
|
Singapore (international) endpoint; use a region-specific general-purpose pay-as-you-go key, not a Token Plan or Coding Plan key. Call the exact IDs …Singapore (international) endpoint; use a region-specific general-purpose pay-as-you-go key, not a Token Plan or Coding Plan key. Call the exact IDs listed here. Enable Free Quota Only per model and wait for the setting to take effect before calling. Check each ID's remaining free token quota before use; removed grants survive only for existing recipients. These API-key profiles do not use the separate OAuth allowance. |
| BazaarLink |
https://api.bazaarlink.ai/v1
|
BAZAARLINK_API_KEY
|
The listed :free routes use the shared free allowance. auto:free selects a free model and can use paid routing on a funded account "unless paid fallback is disabled". Glyph Cluster is a stealth model with an undisclosed identity. |
| Inferbase |
https://api.inferbase.ai/v1
Anthropic format:
https://api.inferbase.ai |
INFERBASE_API_KEY
|
Create an account and a gateway API key. The auto ID routes among models eligible under the key policy; the response names the model that served the …Create an account and a gateway API key. The auto ID routes among models eligible under the key policy; the response names the model that served the request. Managed usage spends a shared budget, so no model is independently free. The vendor documents OpenAI chat and Anthropic Messages tool loops; authenticated generation has not been checked here. |
| Nous Portal (Hermes Agent) |
https://inference-api.nousresearch.com/v1
|
NOUS_PORTAL_API_KEY
|
Select a :free model and use your Nous Portal key; paid variants are outside the free plan. Beam has a 262K context and always reasons; its default effort is medium. |
| IteraCompute |
https://api.iteracompute.com/v1
|
ITERACOMPUTE_API_KEY
|
The documented no-recharge allowance covers qwen/qwen3.8-27b. Eligibility and independent allowances for other routes are unconfirmed; do not apply …The documented no-recharge allowance covers qwen/qwen3.8-27b. Eligibility and independent allowances for other routes are unconfirmed; do not apply Qwen limits to them. Create a key scoped to Qwen and use Chat Completions. Authenticated generation has not been checked here. |
| Regolo AI |
https://api.regolo.ai/v1
|
REGOLO_API_KEY
|
GET /v1/models is public and publishes IDs without prices; use pricing for trial eligibility. brick-v1-beta is a zero-priced prompt classifier and is …GET /v1/models is public and publishes IDs without prices; use pricing for trial eligibility. brick-v1-beta is a zero-priced prompt classifier and is excluded from coding configs. Pricing lists brick-complexity-pro as Core-included, but the public catalog omitted it on 2026-10-31; confirm this router with your account before using it. |
| Token Harbor |
https://tokenharbor.ai/v1
Anthropic format:
https://tokenharbor.aiCodex CLI:
codex -p token-harbor with its profile |
TOKEN_HARBOR_API_KEY
|
The :free suffix selects the free route, which has to be switched on in the dashboard first. /v1/models needs a key; use the IDs listed under Free on …The :free suffix selects the free route, which has to be switched on in the dashboard first. /v1/models needs a key; use the IDs listed under Free on the Models page. For Claude Code the docs say "a free account works" and "every gateway model works on both endpoints". |
| TokenRouter (PaleBlueDot) |
https://api.tokenrouter.com/v1
|
TOKENROUTER_API_KEY
|
Use the zero-priced free-model ID in the default group. PaleBlueDot AI runs tokenrouter.com; similarly named gateways on other TLDs are separate services and their keys do not work here. |
| Mistral Studio |
https://api.mistral.ai/v1
|
MISTRAL_API_KEY
|
Free mode is on by default for a new account; create the key under Studio › API Keys. A key made under Code › Vibe CLI is a plan key that spends the …Free mode is on by default for a new account; create the key under Studio › API Keys. A key made under Code › Vibe CLI is a plan key that spends the plan's Vibe budget, which Mistral tells API users to avoid. The ids are the ones Mistral's own API quickstart and docs call |
| VLM Run Gateway |
https://gateway.vlm.run/v1/openai
|
none — the lane is anonymous |
Omit Authorization; generic bearer tokens return 401. The OpenCode profile uses the catalog context/output limits and a separate 4,096-token request …Omit Authorization; generic bearer tokens return 401. The OpenCode profile uses the catalog context/output limits and a separate 4,096-token request budget for qwen/qwen3.8-27b. Only that ID advertises tool support; the other listed IDs are ordinary-chat choices. |
| Cohere (trial keys) |
https://api.cohere.com/compatibility/v1
|
COHERE_API_KEY
|
Use the OpenAI-compatible endpoint; the native API is https://api.cohere.com/v2. The exact IDs come from the vendor's model pages. north-mini-code-1-0 is free on both Trial and production keys; the other listed models use the Trial allowance. |
| Upstage (Solar API) |
https://api.upstage.ai/v1
|
UPSTAGE_API_KEY
|
— |
| Opper |
https://api.opper.ai/v3/compat
Anthropic format:
https://api.opper.ai/v3/compatCodex CLI:
codex -p opper with its profile |
OPPER_API_KEY
|
The signup guide advertises gemini/gemma-4-31b as free before adding a card, but the model directory omits that route; availability is uncertain. For Claude Code, use ANTHROPIC_BASE_URL=https://api.opper.ai/v3/compat; the client appends /v1/messages. |
| SimpleLLM |
https://api.simplellm.eu/v1
|
SIMPLELLM_API_KEY
|
Use the exact free-prefixed IDs. The public catalog gives both ordinary token prices in SimpleCredits; enabled/status and both prices are rechecked …Use the exact free-prefixed IDs. The public catalog gives both ordinary token prices in SimpleCredits; enabled/status and both prices are rechecked. V1 documents chat completions and SSE; tools depend on the model. Responses is undocumented and the Anthropic adapter is outside the supported V1 contract, so Codex uses the existing LiteLLM bridge. Authenticated generation has not been checked here. |
| MegaNova |
https://api.meganova.ai/v1
|
MEGANOVA_API_KEY
|
Free quota is used first; paid continuation requires enabling the charge switch. Mistral tool execution failed in the tested OpenCode setup, and its LiteLLM-to-Responses path failed in Codex CLI. Manta routers may return tool requests as plain text. |
| SEA-LION (AI Singapore) |
https://api.sea-lion.ai/v1
|
SEA_LION_API_KEY
|
Create a Trial API Key. No expiry or credit balance is published; the documented ceiling is 10 calls/minute. /v1/models requires the key. The free-offer page does not identify which models it serves. |
| OpenTyphoon (SCB 10X) |
https://api.opentyphoon.ai/v1
|
OPENTYPHOON_API_KEY
|
Use typhoon-v2.5-30b-a3b-instruct for chat; it is a Thai fine-tune of Qwen3-30B-A3B. Other catalog routes handle OCR or speech. |
| The Grid |
https://api.thegrid.ai/v1
Anthropic format:
https://messages-beta.api.thegrid.ai |
THE_GRID_API_KEY
|
the ids are instruments from the keyless catalog at api.thegrid.ai/v1/models, where each carries the models it delivered; the Claude Code guide sets ANTHROPIC_BASE_URL=https://messages-beta.api.thegrid.ai, the Messages API in beta |
| Agnes AI |
https://apihub.agnes-ai.com/v1
|
AGNES_AI_API_KEY
|
the quickstart sends its chat completion to apihub.agnes-ai.com/v1/chat/completions with a Bearer key from the Agnes AI Platform dashboard; free and …the quickstart sends its chat completion to apihub.agnes-ai.com/v1/chat/completions with a Bearer key from the Agnes AI Platform dashboard; free and Token Plan keys draw on separate limit pools, and creating more keys of one type does not stack RPM. Agnes AI's China station (agnes-ai.cn) prices the same two Flash models at ¥0 (当前均免费, all currently free) and serves them from api.agnes-ai.cn/v1 |
| Dahl Inference |
https://inference.dahl.global/v1
|
DAHL_INFERENCE_API_KEY
|
Model IDs rotate with network capacity. Allocate tokens from the account pool to each key at /account; a key without allocated tokens answers `402`. |
| Moark (Gitee AI) |
https://api.moark.com/v1
Anthropic format:
https://moark.com/anthropicCodex CLI:
codex -p moark with its profile |
MOARK_API_KEY
|
Use the free experience access token. For Claude Code, set ANTHROPIC_BASE_URL=https://moark.com/anthropic; its guide uses deepseek-v4-flash-0731 in every model slot. The featured models available to the free token are listed in the model square. |
| Sarvam AI |
https://api.sarvam.ai/v1
|
SARVAM_API_KEY
|
The vendor accepts api-subscription-key and Authorization: Bearer. This configuration uses /v1/chat/completions for sarvam-105b; the conversations …The vendor accepts api-subscription-key and Authorization: Bearer. This configuration uses /v1/chat/completions for sarvam-105b; the conversations variant is omitted. The separate /v2/chat/completions and /v2/responses routes are generally available for their listed models. Prices are in rupees. |
| Zro (MoonMath) |
https://zro.moonmath.ai/v1
|
ZRO_API_KEY
|
Create an account and API key for the Free plan and select either listed Flash ID. Other models and auto routing use metered rates. The vendor …Create an account and API key for the Free plan and select either listed Flash ID. Other models and auto routing use metered rates. The vendor documents Chat Completions, Responses and Messages; these configurations use Chat Completions. Authenticated generation and Free activation have not been checked here. |
| Inception Labs (Mercury) |
https://api.inceptionlabs.ai/v1
|
INCEPTION_LABS_API_KEY
|
Mercury 3 supports Chat Completions. Check the authenticated catalog for older Mercury revisions. Mercury Edit 2 uses separate FIM/edit endpoints …Mercury 3 supports Chat Completions. Check the authenticated catalog for older Mercury revisions. Mercury Edit 2 uses separate FIM/edit endpoints; Mercury Voice requires enterprise enablement and Mercury Decide uses a decisions endpoint. The signup token balance is shared across models. |
| NovAI |
https://aiapi-pro.com/v1
|
NOVAI_API_KEY
|
Use a NovAI API key and an exact chat model ID. GLM-4.7-Flash documents OpenAI-compatible function calling and coding-agent setup. CogView and CogVideo are media generators, separate from the chat configurations. |
| Infron |
https://llm.onerouter.pro/v1
|
INFRON_API_KEY
|
Use exact :free IDs with an Infron account key; the OneRouter API domain is documented by Infron. Their daily request pool is shared, and the …Use exact :free IDs with an Infron account key; the OneRouter API domain is documented by Infron. Their daily request pool is shared, and the required balance is held rather than spent on free calls. Catalog availability is public; authenticated generation and account eligibility have not been checked here. |
| FreeInference (Harvard SEAS) |
https://freeinference.org/v1
Anthropic format:
https://freeinference.org/anthropic |
FREEINFERENCE_API_KEY
|
Use a model marked Free in the documentation and returned for your key by authenticated /v1/models. The public view omits qwen3.6-35b and minimax-m3 …Use a model marked Free in the documentation and returned for your key by authenticated /v1/models. The public view omits qwen3.6-35b and minimax-m3 although the docs mark both Free; confirm their availability for your key. Catalog prices are upstream accounting values, not fees charged to Free users. |
| HPC-AI Model APIs |
https://api.hpc-ai.com/inference/v1
Codex CLI:
codex -p hpc-ai with its profile |
HPC_AI_API_KEY
|
ids are the keyless model list's at www.hpc-ai.com/api/maas/v1/models, 2026-09-17, the quick start's own example being moonshotai/kimi-k2.7-code |
| Fireworks AI |
https://api.fireworks.ai/inference/v1
Anthropic format:
https://api.fireworks.ai/inferenceCodex CLI:
codex -p fireworks-ai with its profile |
FIREWORKS_AI_API_KEY
|
— |
| Ollama Cloud |
https://ollama.com/v1
|
OLLAMA_CLOUD_API_KEY
|
An Ollama cloud API key is required. Confirm current free entitlement and select an eligible model in your account before making requests. |
| RouterPlex |
https://api.routerplex.com/v1
Anthropic format:
https://api.routerplex.com |
ROUTERPLEX_API_KEY
|
— |
| Wallaby |
https://api.wallabytoken.com/v1
|
WALLABY_API_KEY
|
Create an account, verify your email and create a token under Tokens → Add token. The official API docs name kimi-k3, support streaming, function …Create an account, verify your email and create a token under Tokens → Add token. The official API docs name kimi-k3, support streaming, function tools and structured outputs. The catalog requires a key; the signup credit is a spending balance, not an individually free model. |
| abliteration.ai |
https://api.abliteration.ai/v1
Anthropic format:
https://api.abliteration.aiCodex CLI:
codex -p abliteration-ai with its profile |
ABLITERATION_AI_API_KEY
|
Abliterated Large Flash and abliterated-model support image inputs; Large v2 and the deprecated Large model are text-only. The deprecated …Abliterated Large Flash and abliterated-model support image inputs; Large v2 and the deprecated Large model are text-only. The deprecated abliterated-model-large remains documented as available for existing integrations. All listed models support Chat Completions, Responses and Messages. |
| Impossibl |
https://api.impossibl.com/v1
Anthropic format:
https://api.impossibl.com |
IMPOSSIBL_API_KEY
|
ids are the keyless catalog's at api.impossibl.com/v1/models, 2026-09-17; which of them the promotional credit can call is not published. The Claude …ids are the keyless catalog's at api.impossibl.com/v1/models, 2026-09-17; which of them the promotional credit can call is not published. The Claude Code guide sets ANTHROPIC_BASE_URL=https://api.impossibl.com and clears ANTHROPIC_API_KEY so the bearer token is used |
| OVHcloud AI Endpoints |
https://oai.endpoints.kepler.ai.cloud.ovh.net/v1
|
none — the lane is anonymous | Use anonymous requests for the free lane; a 429 indicates a rate limit, not the end of the offer. API access keys require a Public Cloud project with a payment method and are billed per token. |
| Standard Compute |
https://api.stdcmpt.com/v1
Anthropic format:
https://api.stdcmpt.com |
STANDARDCOMPUTE_API_KEY
|
one key, two wires: https://api.stdcmpt.com/v1 for OpenAI-shaped clients and https://api.stdcmpt.com for Claude Code — "Claude Code uses …one key, two wires: https://api.stdcmpt.com/v1 for OpenAI-shaped clients and https://api.stdcmpt.com for Claude Code — "Claude Code uses https://api.stdcmpt.com without /v1". /v1/models is keyless and lists the pool the router picks from, 40 unpriced ids on 2026-09-07 with Claude, GPT and Grok among them; requests are smart-routed across it unless a call pins one id. The two ids here are the router's own |
| Novita AI |
https://api.novita.ai/openai
|
NOVITA_API_KEY
|
— |
| Experiential Labs |
https://api.experientiallabs.ai/v1
|
EXPERIENTIAL_LABS_API_KEY
|
An API key is required. There is no ready-to-use free chat ID listed while the pricing-page core offers conflict with the model catalog. The earlier …An API key is required. There is no ready-to-use free chat ID listed while the pricing-page core offers conflict with the model catalog. The earlier Astra and Fable :free promotions have ended; plain model IDs can spend paid credit. Confirm account-specific eligibility before calling a model. |
| Sail Research |
https://api.sailresearch.com/v1
|
SAIL_RESEARCH_API_KEY
|
Select an ID from the current models page and confirm availability with authenticated /v1/models. The vendor documents Responses, Chat Completions …Select an ID from the current models page and confirm availability with authenticated /v1/models. The vendor documents Responses, Chat Completions and Messages APIs. A balanced or flex completion window trades latency for lower token prices; some models are flex-only. The listed examples use the default window. |
| Pollinations.AI |
https://text.pollinations.ai/openai
|
none — the lane is anonymous | The legacy endpoint accepts requests without a key. The replacement gen.pollinations.ai requires personal keys and Pollen credits, bought or earned from Quests. |
| Aion Labs |
https://api.aionlabs.ai/v1
|
AION_LABS_API_KEY
|
Create an API key in the dashboard and send it as a bearer token with the OpenAI Chat Completions format. |
| Arli AI |
https://api.arliai.com/v1
|
ARLI_AI_API_KEY
|
Choose an exact ID from the model catalog. The coding quickstart covers Roo Code and Kilo Code; set the context window for the selected model. Chat completions require an API key. |
| Bytez |
https://api.bytez.com/models/v2/openai/v1
|
BYTEZ_API_KEY
|
model ids are Hugging Face ids, Qwen/Qwen3-4B being the docs' example; the OpenAI SDK examples pass the key as the api key and the docs' curl sends …model ids are Hugging Face ids, Qwen/Qwen3-4B being the docs' example; the OpenAI SDK examples pass the key as the api key and the docs' curl sends it bare in Authorization. /v1/models answers 401 without a key, and closed-source models need a provider key of your own |
| Mixlayer |
https://models.mixlayer.ai/v1
Codex CLI:
codex -p mixlayer with its profile |
MIXLAYER_API_KEY
|
— |
| ModelScope API-Inference (Alibaba) |
https://api-inference.modelscope.cn/v1
Anthropic format:
https://api-inference.modelscope.cn |
MODELSCOPE_API_KEY
|
the key is a ModelScope access token, and it calls only once the account is bound to an Alibaba Cloud account that has passed real-name verification. The ids are the coding-capable rows of the keyless catalog; the Anthropic-format route is in beta |
| SenseNova (SenseTime 商汤) |
https://token.sensenova.cn/v1
|
SENSENOVA_API_KEY
|
An API ID for SenseNova U1 Fast has not been confirmed. |
| Tencent Cloud TokenHub |
https://tokenhub.tencentmaas.com/v1
Anthropic format:
https://tokenhub.tencentmaas.com |
TENCENT_TOKENHUB_API_KEY
|
Use the IDs in the models page's 调用参数 column; /v1/models requires a key. For Claude Code, set ANTHROPIC_BASE_URL=https://tokenhub.tencentmaas.com and ANTHROPIC_MODEL=hy3. |
| SiliconFlow (China) |
https://api.siliconflow.cn/v1
Anthropic format:
https://api.siliconflow.cn |
SILICONFLOW_CN_API_KEY
|
ids are the price list's own; Hunyuan-MT-7B, the other chat row at ¥0, is a translation model. /v1/models needs a key, and the Messages route the API reference documents is https://api.siliconflow.cn/v1/messages |
Ready-made files, regenerated on every update
| configs/opencode.json | Drop-in opencode config with every provider wired up — keys via {env:…}, keyless endpoints work immediately. |
|---|---|
| configs/free-llm.env.example | Commented env exports for any OpenAI-compatible tool. |
| configs/claude-code.sh | One shell function per gateway that serves the Anthropic Messages format — source it and claude-openrouter-free runs Claude Code on that lane. |
| configs/litellm.yaml | LiteLLM proxy for compatible free models; free/nokey switch lanes when one runs out. Start with env -u OPENAI_API_KEY LITELLM_DANGEROUSLY_PERMIT_WEAK_OR_UNSET_MASTER_KEY=true litellm --config configs/litellm.yaml --host 127.0.0.1 to bind to loopback and prevent fallback to your OPENAI_API_KEY. No proxy master key is configured. |
| configs/codex/litellm.config.toml | Codex CLI profile: LiteLLM converts chat completions to the Responses API required by Codex. Run the proxy, copy the file to ~/.codex/, then codex -p litellm (Codex CLI 0.134 and LiteLLM 1.98 or later). |
| llms.txt | The whole list as one text file for LLM search and agents — every live offer with its page, what it needs and where it answers. |
| index.json | The machine-readable registry: every row with its connection details, models, limits and verdicts. |
| browse.html | The same rows as a filterable table — category, no card, no key, OpenAI-compatible, Claude Code route, strong models. |
| feed.xml | Atom feed of every change: rows arriving, rows leaving, free-model lists moving. |
| providers/ | One page per row — the offer in the vendor's own words, the connection details, the evidence the probe reads and the row's history. |
📡 How this list stays fresh
A scheduled job calls every provider twice a week and rewrites this page from what it finds — there is no hand-maintained copy of any figure on it.
Live verification
Every row anchors on something that dies with the offer: a model id in the
vendor's own /models, or wording on the page that states the free tier. A row
that fails 3 probes in a row, or goes 60
days without a pass, drops to the Archive.
The vendor's own words
Limits link to vendor sources. Quotes are checked against those sources; mismatches require review.
Legal free only
Free is by the vendor's own account — a catalog flag, a published price of zero or the vendor's own words on its pricing page. Anything running on pooled consumer accounts, reverse-engineered clients or spoofed apps is refused, and the reason is published with it.
257 services were checked and are not listed — each with the reason on the day it was read and what would change the answer: the whole list of them. Nothing there is disqualified, and every verdict expires after 90 days. Rules, ordering and how to send a row: CONTRIBUTING. A 👁 tag means what you send may be used to train models, by the vendor's own account; the row's page quotes it.A 🧪 new row was added recently on fresh evidence and is provisional until two weeks of probes confirm it.
📦 Archive
Previously listed offers and why they left. Their pages and history are retained.
| Tool | Why it left | Left |
|---|---|---|
| LLM Tech | vendor-announced shutdown on 2026-11-01 | |
| Haimaker | documented free promotions ended on 2026-11-01 | |
| Poolside Platform | vendor-announced shutdown on 2026-10-31 | |
| Amazon Q Developer | delisted on 2026-10-26: closed to new Free accounts and new subscriptions; existing IDE users retain access until the announced end of support | |
| uncloseai (unturf) | delisted on 2026-10-22: public beta closed; the endpoint requires a key and key registration is closed | |
| Yolo-Auto | delisted on 2026-10-15: free accounts now include 0 requests/week and API inference requires a paid plan | |
| Crush + Charm Hyper | delisted on 2026-10-15: The current FAQ retires the monthly free Hypercredit allowance; access requires a paid plan or prepaid credits from $5 (https://hyper.charm.land/faq). | |
| Blue Claw Network | delisted on 2026-10-15: Current onboarding requires purchasing prepaid credits before API use; the previous $5 welcome grant is no longer advertised (https://blueclaw.network/quickstart). | |
| Hugging Face Inference Providers | delisted on 2026-10-08: free accounts no longer receive inference credits — the current billing table lists None and requires purchasing credits or a paid subscription | |
| SambaNova Cloud |
delisted on 2026-09-23: new accounts get no free usage — the console's Free plan has read "Add a payment method and purchase credits to run your …delisted on 2026-09-23: new accounts get no free usage — the console's Free plan has read "Add a payment method and purchase credits to run your first requests" since 2026-08, where until 2026-06-20 it read "free API credits. No credit card required"; only the docs still describe a free tier |
|
| Scaleway Generative APIs | delisted on 2026-09-16: 1,000,000 free tokens once per customer, and "Ordering Scaleway resources requires a valid credit card" — a one-off credit behind a card, which CONTRIBUTING does not admit | |
| Kenari | delisted on 2026-09-16: rejected for cause — the operator's own JavaScript bundle showed its capacity coming from pooled ChatGPT and Codex OAuth credentials, captcha solvers and a proxy pool that multiplies per-IP free quotas | |
| Infomaniak AI Services | delisted on 2026-09-16: a one-month wallet of a million credits, and "A credit card is required to start using the API" — a one-off credit behind a card, which CONTRIBUTING does not admit | |
| Cerebras Inference | delisted on 2026-09-16: its only free offer was $5 of credit granted after a verified payment method is added, expiring in 30 days — a one-off credit behind a card, which CONTRIBUTING does not admit | |
| LongCat API Platform | delisted on 2026-08-14: the daily free quota was gone before the row was listed — the change log retired the Flash line on 2026-05-29 and switched billing on for LongCat-2.0 on 2026-06-30, and no LongCat page still published the 100K tokens a day | |
| Reka AI | delisted on 2026-08-11: the $10 of free credits a month rested on one dated announcement post; no live Reka page corroborated a recurring grant, and the platform sells pay-as-you-go credits | |
| AI21 Labs (Jamba) | vendor-announced shutdown on 2026-08-09 | |
| GitHub Models | vendor-announced shutdown on 2026-07-30 | |
| DeepSeek Platform | delisted on 2026-07-27: the 5M-token signup grant was only ever reported by third parties, and DeepSeek's own pricing page prices every model and names no free grant | |
| MiMo Code | vendor-announced shutdown on 2026-07-26 | |
| Puter.js (free LLM API) | delisted on 2026-07-19: a browser SDK billed to each end user's Puter account, with no HTTP endpoint a coding agent could call; the OpenAI-compatible endpoint read on 2026-09-14 needs a paid plan | |
| Easy GonkaAI API | delisted on 2026-07-19: rejected for cause — its /for-agents page carried a prompt injection aimed at AI agents (auto-signup, credential exfiltration, referral spam) | |
| Aider | delisted on 2026-07-19: BYOK only — the tool is free and open source but bundles no model access of its own, and its docs send readers to other vendors' free tiers, which this list carries directly |
🕰 What changed
The last 8 registry events. The rest is an Atom feed and an append-only history.jsonl.
- 🔄 Free models changed AIHubMix (free models) — dropped laguna-s-2.1, laguna-xs-2.1
- 🔄 Free models changed Kilo Code — dropped laguna-s-2.1, laguna-xs-2.1
- 🔄 Free models changed Nous Portal (Hermes Agent) — dropped laguna-s-2.1, laguna-xs-2.1
- 🔄 Free models changed OpenRouter (free models) — dropped laguna-s-2.1, laguna-xs-2.1
- 🔄 Free models changed Routeway — dropped deepseek-v4-flash, minimax-m2.7
- 📦 Archived Haimaker — documented free promotions ended on 2026-11-01
- ➕ Added Infron — OpenAI-compatible model gateway with a shared recurring Free request allowance, unlocked by a retained prepaid balance
- 📦 Archived LLM Tech — vendor-announced shutdown on 2026-11-01