comparisons
Free Tier AI APIs Every Developer Should Know About: 10 Checked, Only a Few Truly Free
Gemini, OpenRouter, Cloudflare Workers AI, Groq, Mistral, Cohere, Hugging Face, Pollinations, Cerebras and Together AI sorted by what you really get without paying. Some free tiers are real. Several are trials with a card attached. Checked in October 2026.
Every AI provider says "start for free". Open the pricing page and the free part might be a permanent tier, a one time credit that expires in 30 days, a trial key you are not allowed to ship, or nothing at all. If you are building an app that calls a model, that difference decides whether you have a product or a surprise bill.
I checked ten providers and sorted them by what you actually get without paying. Limits in this space move fast, so treat the numbers as a snapshot from October 2026 and confirm in your own dashboard before you build on them.
Real free tiers you can build on
Google Gemini API
The free plan covers input and output tokens on selected models, plus Google AI Studio. The free list includes Gemini 3.8 Flash, Gemini 3.5 Flash-Lite, Gemini 3.1 Flash-Lite, Gemini 3 Flash Preview, Gemini 2.5 Pro, Gemini Embedding 2 and Gemma 4. Some newer models, such as Gemini 3.1 Pro Preview, are marked not available on the free tier. Code execution, URL context and file search are free too. The tradeoff is privacy: on the free tier, your content can be used to improve Google products, and the paid tier turns that off. Rate limits apply per project and are shown in AI Studio, not on the pricing page. See Google Gemini API or visit ai.google.dev.
OpenRouter
One API for many models, with 25 plus free models from 4 free providers. Free models are limited to 20 requests a minute and 50 requests a day. Buy at least $10 of credits and the daily cap rises to 1,000. Failed attempts count toward the cap, and a negative balance blocks free models. Trustpilot reviews are mostly negative, with reports of expired credits and restricted accounts, so test before you depend on it. See OpenRouter or visit openrouter.ai.
Cloudflare Workers AI
10,000 Neurons a day on both the free and paid Workers plans. By the page's own rates, that is roughly 550,000 output tokens on Llama 3.2 1B, or 330,000 on the 3B model. On the free plan, anything past the allocation fails with an error until the next day. A Reddit poster called it fine for experiments and weak for real work. See Cloudflare Workers AI or visit developers.cloudflare.com.
Groq
A free plan exists, with lower limits than the Developer plan, and the exact numbers for your organization show on the console limits page. Batch and Flex processing are Developer plan features. One Trustpilot reviewer praised the free speech to text and another warned that models can be retired with a few weeks notice, so keep your model name in a config file. See Groq or visit groq.com.
Free, with a catch that matters
Mistral AI
The docs describe a free mode, but the limits are not published and point you to a Limits page in the admin panel. In October 2026, free organizations reported a banner asking them to upgrade to activate API keys, plus 429 errors on some models. A Reddit thread asks whether the free tier was removed. Check your own console before planning around it. See Mistral AI or visit mistral.ai.
Cohere
The trial key is free and rate limited, with 1,000 API calls a month. The pricing FAQ says trial keys cannot be used for production or commercial purposes, so this is for learning and prototypes only. See Cohere or visit cohere.com.
Hugging Face
A free Hub account lets you host models, datasets and Spaces, with 100 GB of private storage. Free users get no monthly Inference Providers credits, per the docs, and buy credits to call providers at the providers' own rates. PRO is $9 a month and includes $2 of inference credits. Good for hosting and demos, not a free model API. See Hugging Face or visit huggingface.co.
Pollinations.AI
Free credits come from completing GitHub Quests, trying models or building an app. After that it is pay as you go, where 1 pollen equals $1. The catalog is large, with image, text, audio and video models, and the source is MIT licensed. The free amount is hard to pin down, since there is no fixed weekly allowance. See Pollinations.AI or visit pollinations.ai.
Trials and paid only
Cerebras Inference
A one time $5 credit after you add a payment method, valid for 30 days. Trial limits are 5 requests a minute and 1 million tokens a day per model, on gpt-oss-120b and a Qwen model. When the credit runs out, access pauses until you buy more. There is no permanent free tier. See Cerebras Inference or visit cerebras.ai.
Together AI
No free inference allowance is listed. A G2 reviewer said new accounts start read only until you make an initial deposit. A few models are priced at $0.00, and cheap ones start around $0.14 per million input tokens. Fine as a paid backup, wrong for a zero dollar start. See Together AI or visit together.ai.
What I would use
- A prototype with real traffic and a budget of zero: Gemini API, with the data use tradeoff in mind.
- Many models behind one key while you experiment: OpenRouter.
- Small jobs running at the edge: Workers AI.
- Demos and hosting your own model: Hugging Face.
- A commercial product: budget for paid usage from the start, and use free tiers to prototype.
Whatever you pick, keep the model name and base URL in config and write one small function that wraps the call. When a provider changes its free tier, which happens often, you will swap a single line.