Several AI providers let you call good models through an API without paying, but “free” means different things: a permanent free tier, a small daily allowance, or a trial that ends. This is what each one offers as of 29 October 2026, checked against the providers’ own pages.
At a glance
| Provider | What is free | Free limits (official) | Images on the free option | Guide |
|---|---|---|---|---|
| Gemini API | Free tier for Flash and Flash-Lite | Per project, shown in AI Studio; daily count resets at midnight Pacific time | Yes | Get a Gemini key |
| Groq | Free plan | GPT-OSS 120B: 30 requests a minute, 1,000 a day, 200,000 tokens a day | Yes, with Qwen 3.8 27B | Get a Groq key |
| OpenRouter | Models whose ID ends in :free |
20 a minute and 50 a day (1,000 a day after buying 10 credits once) | Yes, with Gemma 4 | Get an OpenRouter key |
| Cloudflare Workers AI | 10,000 Neurons a day | Resets at 00:00 UTC; on the free Workers plan requests stop until then | Yes, with Gemma 4 26B | Get a Workers AI token |
| Mistral | Free mode, on by default, no card | Set per account; see Admin Panel → API → Limits | Yes | Get a Mistral key |
| Cerebras | Trial: $5 of credits for 30 days | 5 requests a minute, 1 million tokens a day; needs a verified payment method | Yes, with Qwen 3.8 27B | Setup guide |
| OpenAI, Claude | Nothing; pay per use | — | — | Setup guides |
Where to start
- For clear, published daily limits: Groq. It states its free limits per model, they are enough for a full day of questions, and Qwen 3.8 27B covers images.
- For Google’s models: Gemini. Gemini 3.5 Flash-Lite, 3.6 Flash and 3.8 Flash are free and all read images.
- For a second opinion: OpenRouter’s free Gemma 4 models. Fifty requests a day is enough to check answers, not to rely on alone.
- If you already use Cloudflare: Workers AI. Its allowance is measured in Neurons, not requests; a short text question uses only a small part of it.
- For a European provider: Mistral’s Free mode.
Cerebras is a trial, not a free tier: the credits expire after 30 days. OpenAI and Claude have no free API at all; a ChatGPT or Claude subscription does not pay for API use.
What is not free (yet)
The newest flagship models are paid only. Gemini 4 Argon and Gemini 3.1 Pro are not on Gemini’s free tier, and the GPT-6 and Claude 5.5 families, Kimi K3.1 Pro and DeepSeek V4.1 Flash have no free API option on their makers’ platforms. Which new AI models have a free API tracks the recent releases and the closest free alternative for each.
Things that catch people out
- Limits belong to the account or project, not the key. A second key does not double the allowance on Gemini, Groq or OpenRouter.
- Failed requests can count. On OpenRouter an attempt that errors still uses one of the 50 daily free requests.
- Free tiers can use your data. On Gemini’s free tier, Google may use prompts and answers to improve its products; the paid tier does not.
- Text access is not image access. Check that the model you pick accepts images before sending a screenshot of a diagram.
Using a free API in LuciQA
LuciQA is a free desktop app that answers questions about what is on your screen with the provider you choose. The free version connects one provider per question, with no LuciQA question limit; the provider’s own limits still apply.
For clear text, LuciQA’s OCR reads the selected area on your computer and sends only the recognized text, which keeps requests small. For diagrams and formulas, choose Image mode with a model that accepts images. In Pro, every provider set to Every question receives its own request and uses its own quota, so combining Groq and Gemini gives you two answers to compare without paying for either.
Keep keys out of notes, screenshots and repositories. LuciQA stores them encrypted with the operating system’s secure storage. See the provider comparison and the setup guides for current links, and recheck a provider’s page before depending on a free allowance.