Free AI APIs: who has a free tier
Google’s Gemini API is the only major AI API with a standing free tier on its main models. Claude, Qwen and Cohere give free credit or a trial key to start, and the rest bill from the first token. Every allowance below is quoted from the provider’s own page, checked 4 Nov 2026.
Free AI API access by provider
| Provider | Free? | What you get | The catch | Source |
|---|---|---|---|---|
| Free tier | Gemini 3.8 Flash, Gemini 3.6 Flash, the Flash-Lite models, Gemma 4, the embedding models, and the Gemini text-to-speech and transcription models, within rate limits. | Google may use free-tier prompts and responses to improve its products. Gemini 4 Argon, Gemini 3.1 Pro and the image and video models are paid only. | Source for Gemini: Gemini API pricing, checked 4 Nov 2026 (opens in a new tab)4 Nov 2026 | |
| Free credit to start | New users receive a small amount of free credit to test the API. Enterprise teams can ask Anthropic’s sales team for an extended trial. | The pricing page doesn’t say how much; after that, usage is billed per token in advance. | Source for Claude: Claude API pricing, checked 4 Nov 2026 (opens in a new tab)4 Nov 2026 | |
| Free credit to start | A free quota on many Qwen models in the Singapore region, such as 1 million tokens of Qwen3.8 Max, valid for 90 days from activation. | Singapore deployment only; other regions have no free quota. | Source for Qwen: Alibaba Cloud Model Studio pricing, checked 2 Nov 2026 (opens in a new tab)2 Nov 2026 | |
| Free trial key | Free evaluation (trial) keys, limited to 1,000 API calls a month at lower rate limits. | Trial keys aren’t for production; production keys are paid. | Source for Cohere: Cohere API keys and rate limits, checked 4 Nov 2026 (opens in a new tab)4 Nov 2026 | |
| No free tier | No free allowance. Organisations that let OpenAI use their API data get complimentary daily tokens (up to 1M a day on the GPT-6 models on the Launch and Grow plans), and ChatGPT Pro and Business plans include monthly API credits. | Shared traffic may be used to train OpenAI’s models; plan credits expire at the end of each billing cycle. | Source for OpenAI: OpenAI data sharing and complimentary tokens, checked 2 Nov 2026 (opens in a new tab)2 Nov 2026 | |
| No free tier | No free allowance is listed; usage is billed per token from your topped-up balance. Any granted balance on your account is used first. | — | Source for DeepSeek: DeepSeek pricing, checked 4 Nov 2026 (opens in a new tab)4 Nov 2026 | |
| No free tier | The API pricing page lists no free allowance. Grok Build, xAI’s coding tool, has a free tier of its own. | — | Source for Grok: xAI API pricing, checked 4 Nov 2026 (opens in a new tab)4 Nov 2026 | |
| No free tier | Only Mistral Moderation 2 is free; every other model is billed per token. | — | Source for Mistral: Mistral API pricing, checked 4 Nov 2026 (opens in a new tab)4 Nov 2026 | |
| No free tier | Model inference is billed from the first token. Uploading files and extracting their text is temporarily free. | — | Source for Kimi: Kimi API pricing, checked 4 Nov 2026 (opens in a new tab)4 Nov 2026 | |
| No free tier | The pricing page lists no free allowance. Storing cached input is free for a limited time. | — | Source for GLM: Z.ai pricing, checked 4 Nov 2026 (opens in a new tab)4 Nov 2026 |
We list only what each provider states on its own pricing or documentation page. Providers not shown (such as Amazon, Meta, MiniMax and Nvidia) don’t state a free API allowance on the pages we check. Free tiers change often; each row links to its source.
Free speech and transcription APIs
Speech and audio prices| Company | What’s free | Source |
|---|---|---|
| Deepgram | $200 of free credit, with no expiry, then pay as you go. | Source |
| AssemblyAI | A free tier of up to 185 hours of recorded-audio transcription and up to 333 hours of streaming, with no credit card. | Source |
| Google Gemini | Gemini 3.8 Flash TTS, Gemini 3.8 Flash-Lite TTS and Gemini 3.5 Transcribe are on the free tier, within rate limits. | Source |
Already paying for ChatGPT?
ChatGPT Pro plans include the same amount in monthly API credits as their price ($100, $200 or $500), and Business plans $20 or $100 a seat, up to $500 a workspace. You claim them by linking an API organisation you own. Details on the OpenAI pricing page.
Free on your own hardware
Open-weight models cost nothing to download and only electricity to run. A 24GB graphics card or a Mac with 32GB of memory runs capable models such as Qwen 3.8 27B. See what your hardware can run.
When free isn’t enough: the cheapest paid models
All models| Tier | Cheapest current model | Input / output per 1M |
|---|---|---|
| Frontier | DeepSeek V4 ProDeepSeek | $1.32 / $3.96 |
| Workhorse | MiniMax M3MiniMax | $0.30 / $1.20 |
| Small & fast | Ministral 3 3BMistral AI | $0.10 / $0.10 |
List prices from each provider, checked 13 Oct – 6 Nov 2026. A million tokens is roughly 750,000 words.
Questions
Which AI API is free?
Google’s Gemini API is the only major one with a standing free tier on its main models: Gemini 3.8 Flash, Gemini 3.6 Flash, the Flash-Lite models, Gemma 4, the embedding models, and the Gemini text-to-speech and transcription models, within rate limits. Claude gives free credit to start, Qwen gives free credit to start and Cohere gives a free trial key. The others bill from the first request.
Is the OpenAI API free?
No free tier. No free allowance. Organisations that let OpenAI use their API data get complimentary daily tokens (up to 1M a day on the GPT-6 models on the Launch and Grow plans), and ChatGPT Pro and Business plans include monthly API credits. Shared traffic may be used to train OpenAI’s models; plan credits expire at the end of each billing cycle. (Checked 2 Nov 2026 on OpenAI data sharing and complimentary tokens.)
Is the Claude API free?
Free credit to start. New users receive a small amount of free credit to test the API. Enterprise teams can ask Anthropic’s sales team for an extended trial. The pricing page doesn’t say how much; after that, usage is billed per token in advance. (Checked 4 Nov 2026 on Claude API pricing.)
Is the Gemini API free?
Free tier. Gemini 3.8 Flash, Gemini 3.6 Flash, the Flash-Lite models, Gemma 4, the embedding models, and the Gemini text-to-speech and transcription models, within rate limits. Google may use free-tier prompts and responses to improve its products. Gemini 4 Argon, Gemini 3.1 Pro and the image and video models are paid only. (Checked 4 Nov 2026 on Gemini API pricing.)
Is the DeepSeek API free?
No free tier. No free allowance is listed; usage is billed per token from your topped-up balance. Any granted balance on your account is used first. (Checked 4 Nov 2026 on DeepSeek pricing.)
Is there a free Grok API?
No free tier. The API pricing page lists no free allowance. Grok Build, xAI’s coding tool, has a free tier of its own. (Checked 4 Nov 2026 on xAI API pricing.)
What’s the cheapest way to use an AI API if there’s no free tier?
Pick a small model and use batch processing where you can wait for results. The cheapest current models per 1M input / output tokens are DeepSeek V4 Pro ($1.32 / $3.96, frontier), MiniMax M3 ($0.30 / $1.20, workhorse), Ministral 3 3B ($0.10 / $0.10, small & fast). Most providers also take 50% off batch requests.
Can I run AI models for free on my own computer?
Yes, if your hardware can hold them. Open-weight models such as Qwen, Gemma and gpt-oss download for free and cost only electricity to run; the local LLM calculator shows which ones fit your graphics card or Mac and how fast they’ll run.