Free APIs
Free LLM APIs
| Free limits | ||||
|---|---|---|---|---|
Free tiers you can keep using 8Ongoing free access with rate limits. Fine for prototypes, side projects and evals. | ||||
Gemini Flash and Flash-Lite models (e.g. Gemini 3.8 Flash, 3.1 Flash-Lite), Gemma, embeddings
Catch: Pro models are not on the free tier; daily limits reset at midnight Pacific. | Free input and output tokens on Flash models, with per-model RPM, TPM and RPD limits shown in AI Studio | No | Yes (OpenAI compatibility endpoint) | Yes: free-tier prompts are used to improve Google products |
gpt-oss-120b, gpt-oss-20b, Qwen 3.8 27B, Whisper
Catch: Very fast, but 200K tokens a day runs out quickly with long prompts. | 30 requests/min, 1,000 requests/day, 8K tokens/min, 200K tokens/day on the main models | No | Yes | No: inputs and outputs are not retained by default |
Any model ID ending in :free (the list changes often)
Catch: Free models come and go, and 50 requests a day is little without the $10 top-up. | 20 requests/min; 50 requests/day, or 1,000/day once you have bought $10 of credits | No | Yes | Depends on the upstream provider of each free model |
Kimi K3, DeepSeek V4.1 Flash, Nemotron, Llama, Qwen and other open models
Catch: For prototyping and testing only; production needs an NVIDIA AI Enterprise license. | Free hosted endpoints with a free NVIDIA Developer Program account; no quota published | Not stated | Yes | Not stated |
Mistral Small, Medium, Large, Voxtral, OCR
Catch: Limits are not published; not meant for production. | Free mode with included monthly usage; exact limits shown only after login | Not stated | Not stated | Not stated for the free API |
gpt-oss-20b, gpt-oss-120b, Llama 3.1 8B, Gemma, GLM
Catch: Requests fail once the daily Neurons run out, until you move to Workers Paid. | 10,000 Neurons a day on the Workers Free plan, reset at 00:00 UTC | No | Yes (OpenAI-compatible endpoints) | Not stated |
gpt-oss-120b, Llama 3.3 70B, DeepSeek V3.1 and V3.2
Catch: Only 20 requests a day per model. | 20 requests/min, 20 requests/day and 200K tokens/day per model, with no card linked | No | Not stated | Not stated |
Command A, Command R+, Command R, Embed, Rerank
Catch: Trial keys are for evaluation only, not production or commercial use. | Trial key: 20 requests/min on chat, 1,000 API calls a month | Not stated | Not stated | Not stated |
Free trial credits 1One-time credits that expire. | ||||
gpt-oss-120b, Qwen 3.8 27B
Catch: A trial, not a free tier: credits expire after 30 days. | $5 of credits for 30 days; 5 requests/min, 1M tokens/day per model | Yes, to keep using the API after the trial | Not stated | Not stated |
Run models yourself 1No rate limits beyond your own hardware. | ||||
Any open-weight model you download (Gemma, gpt-oss, Qwen, Llama...)
Catch: Speed and model size depend on your GPU and RAM. | Unlimited on your own hardware; free account adds small cloud usage on starter models | No | Yes | No: runs locally; Ollama says cloud prompts are not logged or trained on |
How to get a free Gemini API key
- Go to Google AI Studio (aistudio.google.com) and sign in with a Google account.
- Accept the terms. New users get a Google Cloud project and an API key created automatically.
- To make another key, open aistudio.google.com/apikey and click Create API key.
- Already use Google Cloud? Import your project first under Dashboard, Projects, Import projects.
- Set it as GEMINI_API_KEY in your environment. Keep it out of client-side code and Git.
The free tier covers Flash and Flash-Lite models (e.g. Gemini 3.8 Flash, 3.6 Flash, 3.1 Flash-Lite). Pro models are paid only. Your exact RPM, TPM and RPD limits are listed in AI Studio under Rate limit; daily quotas reset at midnight Pacific.
official source · checked 2026-10-14
How to get a free NVIDIA API key
- Go to build.nvidia.com and create or sign in to an NVIDIA account.
- Join the free NVIDIA Developer Program when asked; it unlocks the hosted endpoints.
- Open any model page, or build.nvidia.com/settings/api-keys, and generate an API key.
- Copy the code sample from the model page: it uses an OpenAI-style endpoint, so the OpenAI SDK works with a changed base URL.
NVIDIA publishes no fixed quota for the free endpoints. They are meant for prototyping and testing; production use needs an NVIDIA AI Enterprise license.
official source · checked 2026-10-14
More: free AI and ML APIs · AI model API pricing · AI coding agents compared · free AI credits for startups.
FAQ
Which LLM API is free?+
Google Gemini API, Groq, OpenRouter, NVIDIA API Catalog, Mistral AI, Cloudflare Workers AI, SambaNova Cloud, Cohere all have free tiers with rate limits. Ollama is free and unlimited if you run models on your own machine.
Is the Gemini API free?+
Yes, for Flash and Flash-Lite models: input and output tokens are free within per-model rate limits, with no credit card. Pro models are paid only. On the free tier, Google may use your prompts to improve its products; the paid tier does not.
How do I get a free Gemini API key?+
Sign in to Google AI Studio (aistudio.google.com) with a Google account, accept the terms, and copy the key it creates, or click Create API key on aistudio.google.com/apikey. No credit card is needed.
Is the NVIDIA API free?+
The hosted models on build.nvidia.com are free with an NVIDIA Developer Program account, for prototyping and testing. NVIDIA publishes no fixed quota, and production use needs an NVIDIA AI Enterprise license.
Which free LLM API is OpenAI-compatible?+
Groq, OpenRouter, NVIDIA, Gemini (through its compatibility endpoint), Cloudflare Workers AI and Ollama accept OpenAI-style requests, so the OpenAI SDK works with a changed base URL and key.
Which free LLM APIs are gone?+
No longer free: GitHub Models (retired July 30, 2026), Hugging Face Inference Providers (free accounts get no credits) and Together AI (prepaid, $15 minimum).