Free AI API tiers and credits
What each AI model API gives a developer for free today, read on the provider’s own page: free tiers you can keep using within limits, one-time credits for new accounts, and Mindket’s deal posts that are still running. Every row names its source and the day it was checked.
Free tiers
| Provider | What is free | Limits | Who can get it | Source |
|---|---|---|---|---|
| Cloudflare Workers AI | Workers AI includes a free daily allowance of Neurons on the Workers Free plan. | 10,000 Neurons per day, reset daily at 00:00 UTC. Above that you need Workers Paid ($0.011 per 1,000 Neurons). Some models (for example kimi-k2.6 and glm-5.2) need a paid billing method. | Anyone with a Cloudflare account on the Workers Free plan. | Provider’s pageChecked November 10, 2026 |
| Cohere | Cohere gives free trial (evaluation) API keys with limited usage. | Trial keys: 1,000 API calls a month; chat models such as Command A, Command R+ and Command R 20 requests/min each. Production keys are paid. | Anyone who creates a trial key on the API keys page. | Provider’s pageChecked November 10, 2026 |
| Google Gemini API (AI Studio) | The Gemini API has a free tier: input and output tokens are free on selected models, used through Google AI Studio. | Limited access to certain models. The pricing page gives no per-model request or token numbers; AI Studio shows your active limits. Content on the free tier is used to improve Google's products. | Any active project or free trial; some projects need verification (rate-limits page). | Provider’s pageChecked November 10, 2026 |
| Groq | Groq has a Free plan with fixed rate limits on its hosted models. | Free Plan Limits tab: openai/gpt-oss-120b, openai/gpt-oss-20b and qwen/qwen3.8-27b each 30 requests/min, 1K requests/day, 8K tokens/min, 200K tokens/day; whisper-large-v3 and whisper-large-v3-turbo 20 requests/min, 2K requests/day, 7.2K audio seconds/hour, 28.8K audio seconds/day. The page calls these a high-level summary with possible exceptions. | Accounts on the Free plan; moving to the Developer plan needs a payment method (Billing FAQs). | Provider’s pageChecked November 10, 2026 |
| NVIDIA NIM (build.nvidia.com) | NVIDIA Developer Program members get free access to NIM API endpoints for prototyping. | No numbers on the page: the rate of requests varies per model and with concurrent users, and your limits are shown in your build.nvidia.com account. For prototyping, research, development and testing only. FAQ last updated 2025-06-03. | Members of the NVIDIA Developer Program (free to join). | Provider’s pageChecked November 10, 2026 |
| OpenRouter | Free model variants (model IDs ending in :free) can be called at no token cost, within request limits. | Under 10 credits purchased all-time: 20 requests/min and 50 requests/day. At least 10 credits purchased: 20 requests/min and 1,000 requests/day. | Any account; the daily cap depends on credits purchased all-time. The FAQ also says new users get a small free allowance, with no amount stated. | Provider’s pageChecked November 10, 2026 |
| SambaNova Cloud | SambaCloud has a Free Tier with fixed daily limits on its models. | For each of DeepSeek-V3.1, Meta-Llama-3.3-70B-Instruct, gpt-oss-120b, DeepSeek-V3.2 and gemma-4-31B-it: 20 requests/min, 20 requests/day, 200,000 tokens/day. | Accounts with no payment method linked. | Provider’s pageChecked November 10, 2026 |
| SiliconFlow (硅基流动) | SiliconFlow offers a set of free models; calls to them cost nothing. | Free-model calls show 0 cost on the bill. Their rate limits are fixed values; the numbers are not on this page. Paid models use usage-level tiers. | Accounts that have completed real-name verification (实名认证). | Provider’s pageChecked November 10, 2026 |
| Vercel AI Gateway | Every Vercel team gets a monthly included credit for AI Gateway, usable on a subset of free-tier models. | Credit amount not stated on the page. Free tier covers a subset of models with lower per-model rate limits; free credits start at the first request; buying credits moves the team to the paid tier and the monthly free credit no longer applies. | Every Vercel team account; the Getting started page says a valid payment method must be added to use the free credits. | Provider’s pageChecked November 10, 2026 |
| Zhipu / Z.ai (GLM) | Z.ai lists GLM-4.7-Flash and GLM-4.5-Flash as free models. | Input, cached input, cache storage and output are all listed as Free. The pricing page gives no request or token numbers; concurrency limits are set per account. | Not stated on the pricing page. | Provider’s pageChecked November 10, 2026 |
Free credits
| Provider | What is free | Limits | Who can get it | Source |
|---|---|---|---|---|
| Alibaba Cloud Model Studio (International, Singapore) | New Model Studio accounts in the Singapore region get a free quota of tokens per model. | Typically 1,000,000 tokens per model, input and output combined. Valid 90 days from activation, then it expires even if unused. Only Singapore-region models with International deployment scope qualify. After it runs out you are billed pay-as-you-go unless you turn on Free Quota Only. | Accounts activating Model Studio for the first time; account information must be completed first. Re-registering does not grant another quota. | Provider’s pageChecked November 10, 2026 |
| Alibaba Cloud 百炼 (China, Beijing) | New 百炼 accounts get a free quota of tokens per model in the China (Beijing) region. | Typically 1,000,000 tokens per model, independent for each model. Valid 90 days from activation. Only the North China 2 (Beijing) region has a free quota. After it runs out you are billed pay-as-you-go unless you turn on the stop-when-used-up switch. | Accounts activating 百炼 for the first time; real-name verification (实名认证) is required or no quota is issued. | Provider’s pageChecked November 10, 2026 |
| Baidu Qianfan (百度千帆) | New Qianfan users get free tokens on 17 listed services, including ERNIE, DeepSeek, Kimi and Qwen models. | 1,000,000 tokens per listed service, valid 3 months. Covers online inference of preset models only, not batch inference. The page is dated 2025-11-17. | New users: the platform is opened and the quota issued automatically after you read and agree to the user agreement (rule effective from 2025-10-24). | Provider’s pageChecked November 10, 2026 |
| Cerebras | New Cerebras accounts get $5 in free credits that work on all public models. | $5 in credits, expiring 30 days after they are granted. Free Trial limits for gpt-oss-120b and qwen-3.8-27b: 5 requests/min, 30K uncached tokens/min, 90K total tokens/min, 1M tokens/hour, 1M tokens/day. The page says there is no permanently free tier. | New accounts that add a verified payment method; without one, Playground and API access stay inactive. | Provider’s pageChecked November 10, 2026 |
| Tencent Hunyuan (腾讯混元) | First-time Hunyuan text-model activation gives a one-time free resource package of calls. | 100万 (1,000,000) tokens shared across Hunyuan-a13b, Hunyuan-role-latest, Hunyuan-translation, Hunyuan-translation-lite and the Hunyuan vision models, plus 100万 tokens for Hunyuan-embedding. Valid 1 year from activation, then it expires. After it runs out, post-pay does not start automatically. The page says Hunyuan is moving to TokenHub. | Accounts with Tencent Cloud enterprise or personal real-name verification, on first activation. | Provider’s pageChecked November 10, 2026 |
| Volcano Engine Ark (火山方舟) | New Ark users get a free inference quota; in worry-free mode the service stops when it is used up. | 50w (500,000) tokens of free inference quota in worry-free mode. The companion free-quota page (docs/82379/1399514) says it is counted per model and shared under the main account. Covers pay-by-token online inference only, not batch inference, plug-ins or knowledge bases. No end date stated. | Worry-free mode: accounts (real-name verified personal or enterprise) that have never activated a model service. To continue after the quota, real-name verification and manual activation are needed. | Provider’s pageChecked November 10, 2026 |
Deals running
| Deal | Ends |
|---|---|
| Higgsfield's free 3-day trial in Claude: the conditions, how to claim it and what it makes72 hours · free in Claude, then Ultra unless you cancel | No end day statedChecked November 11, 2026 |
| OpenAI's Codex credits for students: who can claim 2,500 free credits?Worth US$100 · by OpenAI’s own count | No end day statedChecked November 4, 2026 |
| Google AI Pro free for students: who gets the free year?$0 · for 12 months | Until December 31, 2026Checked November 4, 2026 |
| Manus's 1,000 free credits: do you have to comment to get them?$0 · no comment or code needed | No end day statedChecked November 4, 2026 |
| Higgsfield's 7-day unlimited video bonus: is it free for new users?0 credits · per 20-second video, on paid plans | No end day statedChecked November 3, 2026 |
Questions
Which AI APIs have a free tier?
As of the day each row was checked: Cloudflare Workers AI, Cohere, Google Gemini API (AI Studio), Groq, NVIDIA NIM (build.nvidia.com), OpenRouter, SambaNova Cloud, SiliconFlow (硅基流动), Vercel AI Gateway and Zhipu / Z.ai (GLM). Each row gives the limits and links the provider’s page that states them.
What is the difference between a free tier and free credits?
A free tier lets you keep using the API at no cost within its limits, such as requests or tokens a minute or a day. Free credits are a one-time amount for a new account: once they are used up or expire, use is billed.
Can I build a product on a free tier?
Read the provider’s terms first: free tiers often have low rate limits, and some providers state different data terms for free use. Each row links the page with its limits.
How current is this list?
Each row shows the day it was checked at its source. Rows not checked for 90 days are taken off, and offers past their stated last day move to Recently ended for 30 days.