Free LLM APIs
VerifiedAs of 29 Oct 2026, 5 providers have an ongoing free LLM API and 6 more offer limited or trial access. Groq publishes the highest free limit (1,000 requests a day on gpt-oss-120b), Cloudflare gives a daily compute budget with no training on your data, and Google's Gemini API offers Gemini 3.8 Flash free but only shows its limits in your console. 4 providers we track, including Hugging Face Inference Providers and GitHub Models, have no free tier now. More categories, such as coding assistants, are coming to free tiers for AI developers.
| Provider | Free limit | Card? | The catch | Checked | Status |
|---|---|---|---|---|---|
| Groq gpt-oss-120b, gpt-oss-20b, Qwen3.8 27B, Whisper | 30 RPM · 1,000/day | Not confirmed | Only 8K tokens a minute, and Llama models moved to Enterprise plans | Free | |
| OpenRouter 14 models tagged :free, including Gemma 4 31B and Nemotron 3 Ultra | 20 RPM · 50/day | No | 50 requests a day unless you have bought $10 of credits (then 1,000 a day) | Free | |
| Google Gemini API Gemini 3.8 Flash, 3.6 Flash, 3.5 and 3.1 Flash-Lite, Gemma 4 | Per model, shown in AI Studio | No | Free prompts can be used to improve Google products, and the free tier is not allowed for apps serving EEA, UK or Swiss users | Free | |
| Cloudflare Workers AI Open models at the edge: Llama 3.x, Qwen3, Granite 4, gpt-oss | 10,000 Neurons/day · 300 RPM | Not confirmed | The daily budget is in Neurons, so real capacity depends on the model and output length | Free | |
| Mistral Mistral Studio API models in Free mode, such as Mistral Small | Shown in the console | No | Prompts can be used for training unless you opt out, and limits are the lowest tier | Free | |
| SambaNova Cloud gpt-oss-120b, DeepSeek V3.1, Llama 3.3 70B | 20 RPM · 20/day | No | Only 20 requests a day per model on the free tier | Limited | |
| Cohere Command A, Command R and R+, Embed, Rerank (trial keys) | 20 RPM · 1,000 calls/month | No | Trial keys are not allowed in production or for commercial use | Limited | |
| NVIDIA NIM (build.nvidia.com) NVIDIA-hosted models: Nemotron, Llama and more | Varies by model, shown in your account | Not confirmed | NVIDIA describes it as for prototyping and testing; production needs NVIDIA AI Enterprise | Limits not published | Limited |
| Cerebras gpt-oss-120b, Qwen 3.8 27B (trial credits) | $5 credit for 30 days · 5 RPM | Yes | No renewing free tier: the trial credit needs a verified payment method and expires | Limited | |
| OpenAI API GPT-6, GPT-5 and o-series models, paid after a test request | Test requests only | Not confirmed | After a test request, you need at least $5 of prepaid credit to keep calling the API | Limited | |
| Claude API Claude Haiku 5.5, Sonnet 5.5 and Opus 5.5, on a small trial credit | Small trial credit | Not confirmed | Anthropic does not say how big the credit is, and once it runs out you have to buy credits before any call | Limited | |
| Hugging Face Inference Providers 200+ models routed to partner providers | No free credits for free accounts | Yes | Free accounts now have to buy credits; PRO includes $2 a month | Not free | |
| GitHub Models Retired | Shut down 30 Jul 2026 | n/a | The playground, API and bring-your-own-key were shut down for everyone | Not free | |
| Together AI No free models | No free trial | Yes | Together does not offer free trials; the first purchase is at least $15 | Not free | |
| DeepSeek API No free models | No free tier or trial | n/a | DeepSeek publishes no free credit for new accounts, so you pay from a prepaid balance | Not free |
Every row links to the provider's own docs on its profile page. "Not confirmed" means the provider does not publish that detail. How we verify
Frequently asked
Which free LLM API has the highest limits?
Of the providers that publish fixed numbers, Groq allows the most requests a day without buying anything: 1,000 a day (30 a minute) on gpt-oss-120b. OpenRouter also reaches 1,000 a day once you have bought $10 of credits, and Cloudflare allows more requests a minute within its daily compute budget. Google's Gemini API and Mistral also have free tiers, but they only show their limits inside your account console. Checked 29 Oct 2026.
Do free LLM APIs need a credit card?
Most free tiers here do not. Google Gemini API, OpenRouter, Mistral, Cohere, SambaNova Cloud say no card is needed. Cerebras now asks for a verified payment method before it gives trial credits, and several former free options now require buying credits.
Can I use a free LLM API in a commercial product?
Sometimes. Cohere's trial keys are explicitly not allowed for production or commercial use, and NVIDIA describes its hosted endpoints as for prototyping. Google allows commercial use of the Gemini free tier but may use your prompts to improve its products, and does not allow it for apps serving EEA, UK or Swiss users. Read each provider's terms before you ship.
Which free LLM APIs do not train on my prompts?
Cloudflare Workers AI states in its docs that your data is not used for training. Google's Gemini free tier and Mistral's free mode can use prompts for training (Mistral lets you opt out). For the others it is not stated clearly, so we mark it as not confirmed.