Free API tiers - verified
Free LLM APIs: every free tier, its limits and its data terms
We read the docs, pricing and terms of 17 LLM API providers on October 18, 2026. 8 have a standing free tier, 3 give credits that run out instead, and 6 have no free tier. Of the 11 you can start on without paying, 6 say they do not train on your data, 2 may train on it, 1 depends on the provider serving the model and 2 do not say.
Credits are listed apart from free tiers because they run out. Data terms to read first - Before you build
- Free tiers
- 8
- Credits
- 3
- No free tier
- 6
- Free + no training
- 6
Every provider, side by side
Not stated means the provider's own pages do not say, not that the answer is no. The training column reads each provider's data terms for the free tier or trial; open a provider for the verbatim quote.
17 providers
| Provider | Offer | Card required | Headline limits | Trains on your data?free tier or trial |
|---|---|---|---|---|
| No providers match these filters. | ||||
| Google Gemini API | Free tiera standing Free Tier, shared with Google AI Studio | Not stated | Free of charge on 12 listed models; per-model RPM, TPM and RPD shown only in Google AI Studio | YesOutside the EEA, Switzerland and the UK |
| Mistral | Free tierFree mode, the default state for new accounts | No | Lowest limits, shown only on the logged-in Limits page | Yes, opt-outOpt-out toggle in the Admin panel |
| Groq | Free tiera standing Free tier until you upgrade to the Developer tier | Not stated | openai/gpt-oss-120b: 30 RPM, 1K RPD, 8K TPM, 200K TPD | NoServices Agreement covers all Cloud Services |
| OpenRouter | Free tierfree model variants whose IDs end in :free | Not stated | 20 requests per minute; 50 requests per day, or 1000 after buying at least 10 credits | Depends on the serving providerSet by the provider serving each free model |
| Z.ai (GLM) | Free tierfree Flash models: GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash | Not stated | 3 Flash models priced Free; rate limits shown only in the logged-in console | NoFor API Services, unless you explicitly agree |
| Cloudflare Workers AI | Free tiera daily free allocation on the Workers Free and Paid plans | Not stated | 10,000 Neurons per day shared across models; 300 requests per minute for text generation | No |
| NVIDIA API Catalog | Free tierfree NIM API access for prototyping through the NVIDIA Developer Program | Not stated | No numeric limits published; trial use may be limited by instances or duration | Not statedTrial terms say content is not stored or used after each session |
| Cohere | Free tierfree trial (evaluation) API keys | Not stated | 1,000 API calls a month; 20 req / min on Command models | Not statedOpt-out toggle exists; no-training commitments cover paying customers |
| Anthropic Claude API | Creditsa small amount of free credits for new users | Not stated | Credit amount not stated; new orgs may start in the Evaluation tier | NoCommercial Terms |
| Alibaba Cloud Model Studio (Qwen) | Creditsa new-user free quota per model, granted when you activate Model Studio in the Singapore region | Not stated | 1 million tokens per model, valid 90 days, Singapore region only | NoUnless you separately consent |
| Cerebras Inference | Creditsone-time Free Trial credits after you add a verified payment method | Yes | $5 one-time, expires 30 days after grant; gpt-oss-120b: 5 RPM, 1M TPD | No |
| OpenAI API | No free tierno free tier for text generation | Not stated | No Free-tier limits on text-generation models; complimentary tokens only for paying tiers that share data | NoUnless you opt in to share data |
| DeepSeek API | No free tierno free tier: the API is prepaid | Not stated | None documented | Not statedThe API terms do not address training; DeepSeek's general privacy policy says Services inputs may train its models |
| Moonshot AI (Kimi) | No free tierno free tier: you recharge at least $1 to start | Not stated | None; recharge at least $1 to start | YesUnless agreed otherwise in writing |
| xAI Grok API | No free tierno free tier: you load credits to start | Not stated | None documented | NoWithout your explicit permission |
| Hugging Face Inference Providers | No free tierno free credits for free accounts | Not stated | None for free accounts; monthly credits only on PRO, Team and Enterprise | Depends on the serving providerHugging Face stores no routed data; each provider's policy applies |
| GitHub Models | No free tierno free tier, because the service is retired | Not stated | Retired as of July 30, 2026 | Not statedService retired |
Three data terms to read first
A free tier is often paid for with your prompts. These three are the ones most likely to surprise you.
Google: unpaid quota prompts can improve Google products
On the Gemini API's unpaid quota and in AI Studio, Google may use your prompts and responses to improve its products and machine learning technologies, and human reviewers may read them. Users in the EEA, Switzerland and the UK get the paid-service data terms instead.
“To help with quality and improve our products, human reviewers may read, annotate, and process your API input and output. Google takes steps to protect your privacy as part of this process. This includes disconnecting this data from your Google Account, API key, and Cloud project before reviewers see or annotate it. Do not submit sensitive, confidential, or personal information to the Unpaid Services.”
ai.google.dev “If you're in the European Economic Area, Switzerland, or the United Kingdom, the terms under "How Google uses Your Data" in "Paid Services" apply to all Services, including Google AI Studio and unpaid quota in the Gemini API, even though they are offered free of charge.”
ai.google.dev Mistral: Free mode trains on input and output by default
Mistral may train on your Free mode input and output unless you opt out in the Admin panel. The API and Vibe opt-out toggles are separate.
“Free mode: As stated during subscription, we may use your data (input and output) to train our artificial intelligence models. You have the right to opt out of this program at any time.”
help.mistral.ai How to opt out (Mistral help center)Mistral free tier details
OpenRouter: the data policy depends on who serves the free model
OpenRouter itself does not store prompts unless you opt in, but each free model is served by a provider with its own policy. 6 of the 12 free models listed here are served by Nvidia and Thinking Machines, which OpenRouter marks "May train". Opting out of training in your account settings stops routing to those providers.
“On your account settings page, you can set whether you would like to allow routing to providers that may train on your data (according to their own policies). There are separate settings for paid and free models.Wherever possible, OpenRouter works with providers to ensure that prompts will not be trained on, but there are exceptions. If you opt out of training in your account settings, OpenRouter will not route to providers that train. This setting has no bearing on OpenRouter's own policies and what we do with your prompts.”
openrouter.ai
Caveats
Read before you build on a free tier
Rate limits change
Free-tier limits are summaries the providers revise without notice, and several publish exact values only in a logged-in console.
“The following is a high level summary and there may be exceptions to these limits. You can view the current, exact rate limits for your organization on the limits page in your account settings.”
Groq - console.groq.com “Free variants provide access to models without cost, but may have different rate limits or availability compared to paid versions.”
OpenRouter - openrouter.ai Trial keys say prototyping only
Trial and evaluation access is for testing, not for serving real users. Plan the upgrade before you launch.
“NIM access through the NVIDIA Developer Program is for prototyping, research, development and testing purposes only and does not include enterprise features or support.”
NVIDIA API Catalog - docs.api.nvidia.com “Developers wanting to test different applications or build proofs of concept can use all of Cohere's models and APIs can do so with a trial key”
Cohere - docs.cohere.com “Free mode (the default) has the lowest limits, intended for evaluation and prototyping.”
Mistral - help.mistral.ai Alibaba bills automatically after the free quota
When an Alibaba Cloud Model Studio free quota runs out, usage moves to pay-as-you-go unless "Free Quota Only" is turned on, and it is off by default.
“After the free quota is exhausted, you are automatically billed on a pay-as-you-go basis. To avoid unexpected charges, enable Free Quota Only in advance so that the service stops when the quota runs out.”
Alibaba Cloud Model Studio (Qwen) - alibabacloud.com “Worry-free mode (Free Quota Only) is disabled by default; enable it in the console if needed.”
Alibaba Cloud Model Studio (Qwen) - alibabacloud.com
FAQ
Which LLM APIs have a free tier in 2026?
Of the 17 providers checked on October 18, 2026, 8 have a standing free tier: Google Gemini API, Mistral, Groq, OpenRouter, Z.ai (GLM), Cloudflare Workers AI, NVIDIA API Catalog and Cohere. Anthropic Claude API, Alibaba Cloud Model Studio (Qwen) and Cerebras Inference give credits instead, which run out. OpenAI API, DeepSeek API, Moonshot AI (Kimi), xAI Grok API, Hugging Face Inference Providers and GitHub Models have no free tier.
Which free LLM APIs do not train on your data?
Groq, Z.ai (GLM) and Cloudflare Workers AI among the free tiers, and Anthropic Claude API, Alibaba Cloud Model Studio (Qwen) and Cerebras Inference among the credit offers, state that they do not use your API data for training. Google Gemini API (outside the EEA, Switzerland and the UK) and Mistral (opt-out toggle in the Admin panel) may train on free-tier data.
Do free LLM APIs need a credit card?
Mistral states that no credit card is required. Cerebras Inference requires a verified payment method before the free credits activate. The other 9 providers with a free tier or credits do not state it either way.
Is the OpenAI API free?
No, not for text generation. Every text-generation model page checked publishes limits only for the paid Build, Launch and Grow tiers. Eligible Build, Launch and Grow organizations that opt in to share inputs and outputs with OpenAI can get daily complimentary tokens; free accounts cannot, because a positive balance is required.
Verified from each provider's own docs, pricing and terms pages. Free tiers, limits and data terms change without notice; check the linked source before you rely on one.
Newsletter
Know when a free tier changes
One email per week. New model releases, pricing and limit changes, and what they mean for what you build.
Keep exploring
AI rate limits and usage limits
Paid-tier limits per provider, copied verbatim from the docs.
Model API pricing and specs
First-party prices per 1M tokens for when the free tier runs out.
OpenAI-compatible API endpoints
Base URLs and auth for every provider's compatibility layer.
Run models locally
Open-weight models that run on your own hardware, free of rate limits.