freellmpool › reference
These 22 providers span recurring free tiers, finite trials, keyless endpoints, and explicit pin-only routes. Three need no API key at all, and LLM7 is key-optional. The table shows whether a key is required, roughly how many routes are enabled, and an access signal where known. Disabled or pending-canary routes are not automatic capacity. The catch with using several is that each has its own SDK and limits — the open-source freellmpool pools the enabled, eligible routes behind one OpenAI-compatible endpoint with failover. Disabled, priced, and finite-trial routes remain visible for explicit decisions but are not all automatic free capacity.
| Provider | API key? | Enabled routes* | Access signal** |
|---|---|---|---|
| Pollinations | No (keyless) | 3 | Anonymous; rate-limited per IP |
| OVHcloud AI Endpoints | No (keyless) | 17 | Anonymous tier (chat + embeddings + Whisper) |
| LLM7 | Optional | 3 | ~60/hour without a token |
| Kilo Gateway | No (keyless) | 16 | Anonymous; ~200 req/hour per IP |
| OpenCode Zen | No (keyless) | 0 | Cataloged routes disabled by default pending opt-in |
| Groq | Yes (free) | 7 | Model-specific free-plan limits; Compound routes are 250 requests/day |
| Cerebras | Yes (trial) | 2 | Finite $5 trial; documented routes are explicit pins, not automatic capacity |
| Google Gemini | Yes (free) | 1 | Current free-tier limits vary by project/model |
| OpenRouter | Yes (free) | 8 | Free-route limits depend on account credits and current policy |
| NVIDIA NIM | Yes (free) | 11 | Account credits and model-specific rate limits |
| Cloudflare Workers AI | Yes (free) | 22 | Daily Neuron allowance |
| Mistral | Yes (free) | 36 | Account- and model-specific experimentation limits |
| Cohere | Yes (free) | 21 | Trial keys, rate-limited |
| SambaNova | Yes (free) | 0 | Catalog routes disabled pending current verification |
| Z.ai / Zhipu GLM | Yes (free) | 2 | ~1,000 req/day |
| Ollama Cloud | Yes (free) | 7 | Qualitative free limits |
| Hugging Face (router) | Yes (free) | 21 | Monthly included Inference credits |
| Aion Labs | Yes (free) | 4 | 20K tokens/day; no card |
| ModelScope API Inference | Yes (free) | 6 | 2,000 calls/day total; dynamic per-model caps |
| Morph | Yes (priced) | 0 | Current aliases are priced and remain disabled |
| Vercel AI Gateway | Yes | 8 | One zero-price automatic route; priced pins excluded; disabled candidates await canaries |
| SiliconFlow | Yes (free) | 1 | Fixed free-model limits; identity verification required |
* Enabled routes across chat, embeddings, and transcription in freellmpool's catalog (196 capability routes in total; 177 are chat routes); varies as tiers change. ** Access signals are approximate and change without notice — always check the provider's current docs. freellmpool tracks your real per-day usage as you go.
pip install freellmpool
freellmpool ask "Explain the CAP theorem in one sentence." # keyless when an eligible route is available
freellmpool providers # see which tiers you've configuredAdd credentials for eligible providers to grow capacity; signup links are in the accounts guide. See also: using multiple free LLM APIs together · best free LLM API gateway.
The catalog distinguishes recurring free tiers, finite trials, keyless endpoints, priced pins, and disabled candidates. Pollinations, OVHcloud, and Kilo Gateway have enabled keyless routes; LLM7 accepts an optional key. Cerebras is finite-trial pin-only, while OpenCode Zen, SambaNova, and Morph currently have no enabled routes. Check each row and the dated activity audit instead of treating all 22 provider groups as interchangeable free capacity.
There is no durable single winner: recurring allowances are model-, account-, and project-specific, while Cerebras now offers only finite trial credit. Check each provider's current limits and pool several recurring tiers so one cap is not a bottleneck.