Free LLM APIs and AI API Credits
Which model APIs you can call without paying today, what each free allowance covers and what the provider does with your prompts, read from the providers' own pages.
Free allowances read on each provider's own pages, all checked on or after . Free tiers and their limits change often: confirm them on the provider's page before you build on them.
AI APIs you can call for free today include Gemini API (free input and output tokens on certain models), Groq (free tier API access within per-model rate limits), OpenRouter (free model variants, IDs ending in :free), Mistral ($10 a month in API credits), and Cloudflare Workers AI (10,000 Neurons a day at no charge). Each of the 14 allowances below links to the provider's page that states it, with its limits and what the provider does with free-tier prompts.
Several model providers and inference hosts let developers call their APIs for free. Some run a standing free tier with rate limits; others give an amount of free credit, once or every month. Both are meant for prototyping and evaluation, and each comes with its own limits, model list and data terms.
The terms that decide whether an allowance fits your use are how many requests or tokens it covers, which models it includes, whether you must add a payment card, whether prompts sent on it may be used to improve the provider's models, and what happens when you outgrow it. Each is quoted below from the provider's own page, with the date we read it.
Which APIs we picked and the order they appear in are our editorial choices; the terms are the providers'. A free allowance is a way to start, not a way to run production. When you need more, the startup programs at the end of this guide give far larger model credits, most of them by application.
Free tiers you can keep using
No end date: the provider lets you call these models for free within the stated limits.
- 1
Gemini API
Prototyping on Gemini models with no billing account
New accounts start on a free tier with free input and output tokens on certain Gemini models, and billing is only needed to move to the paid tier. Google uses free-tier content to improve its products, and apps for users in the EEA, Switzerland or the UK must use the paid tier.
- Your prompts and data
- Used to improve Google products and ML technologies (source: ai.google.dev)
- Beyond the free allowance
- Paid tier needs a billing account and a $5 prepay (source: ai.google.dev)
Terms read on ai.google.dev on Gemini API: the free allowance and how to start Google for Startups Cloud Program: more credits for startups - 2
Groq
Trying open models such as gpt-oss with no card
The free plan gives API access to open models such as gpt-oss-120b and qwen3.8-27b at per-model rate limits, and a payment method is only needed to upgrade. Groq's agreement bars it from using your inputs or outputs for training without your permission.
- Your prompts and data
- Not used for training without your permission (source: console.groq.com)
- Beyond the free allowance
- Developer tier: more capacity, pay-as-you-go pricing (source: console.groq.com)
Terms read on console.groq.com on Groq: the free allowance and how to start - 3
OpenRouter
Trying 25+ free models at low volume
OpenRouter lists 25+ free models whose IDs end in :free, capped at 20 requests a minute and 50 a day, or 1,000 a day once you have bought 10 credits. You choose in your account settings whether free models may route to providers that train on prompts.
- Your prompts and data
- Choose whether free models route to training providers (source: openrouter.ai)
- Beyond the free allowance
- Buy 10 credits for a higher limit, or use paid variants (source: openrouter.ai)
Terms read on openrouter.ai on OpenRouter: the free allowance and how to start OpenRouter for Startups: more credits for startups - 4
Mistral
Evaluating Mistral models with no card
The Free plan includes $10 a month in API credits and needs no credit card, at the lowest rate limits and meant for evaluation and prototyping. Mistral may use your inputs and outputs to train its models on this plan.
- Your prompts and data
- May use your inputs and outputs to train its models (source: help.mistral.ai)
- Beyond the free allowance
- Pay-as-you-go bills per token, or usage can stop (source: docs.mistral.ai)
Terms read on mistral.ai on Mistral: the free allowance and how to start - 5
Cloudflare Workers AI
Light daily use of Workers AI models
Anyone can use 10,000 Neurons a day at no charge, with 300 requests a minute for text generation. Some models need a paid billing method, and Cloudflare says it does not use your content to train models or improve its services without your consent.
- Your prompts and data
- Not used for training or to improve services (source: developers.cloudflare.com)
- Beyond the free allowance
- Workers Paid plan: $0.011 per 1,000 Neurons (source: developers.cloudflare.com)
Terms read on developers.cloudflare.com on Cloudflare Workers AI: the free allowance and how to start Cloudflare for Startups: more credits for startups - 6
NVIDIA API Catalog
Prototyping on NVIDIA-hosted model endpoints
NVIDIA Developer Program members get free NIM API endpoints for prototyping, at up to 40 requests a minute. Inputs and outputs are recorded to improve NVIDIA products and models, and production use needs an NVIDIA AI Enterprise license.
- Your prompts and data
- Inputs and outputs recorded to improve NVIDIA models (source: assets.ngc.nvidia.com)
- Beyond the free allowance
- AI Enterprise license from $4,500 per GPU a year (source: docs.api.nvidia.com)
Terms read on docs.api.nvidia.com on NVIDIA API Catalog: the free allowance and how to start NVIDIA Inception: more credits for startups - 7
Cohere
Evaluating Cohere models at low volume
A free Trial API key comes with a Cohere account and no payment method, limited to 1,000 API calls a month and barred from production or commercial use. Trial inputs and outputs may be used to improve Cohere's models.
- Your prompts and data
- Trial inputs and outputs may be used to improve models (source: cohere.com)
- Beyond the free allowance
- Production keys are billed pay as you go (source: cohere.com)
Terms read on cohere.com on Cohere: the free allowance and how to start - 8
Hugging Face
Trying hosted models with a small monthly credit
Every Hugging Face account gets $0.10 a month in credits for Inference Providers, which reach hundreds of models from partner providers. Hugging Face says it does not store request bodies or use user data for training, and extra usage needs purchased credits.
- Your prompts and data
- No user data stored for training; bodies not stored (source: huggingface.co)
- Beyond the free allowance
- Buy additional credits to keep using the API (source: huggingface.co)
Terms read on huggingface.co on Hugging Face: the free allowance and how to start - 9
Vercel AI Gateway
Light use of a model gateway on a Vercel team
Every Vercel team gets $5 a month in AI Gateway credit on a free tier that covers a subset of models at lower rate limits. A payment method must be on the team, and buying credits moves it to the paid tier and ends the monthly credit.
- Your prompts and data
- Routing ignores providers' training policies by default (source: vercel.com)
- Beyond the free allowance
- Buying credits moves you to paid; free credit ends (source: vercel.com)
Terms read on vercel.com on Vercel AI Gateway: the free allowance and how to start Vercel for Startups: more credits for startups - 10
OpenAI
API teams already on OpenAI that can share traffic
Eligible organizations that opt in to share API traffic with OpenAI get daily complimentary tokens, 1M or 10M by model group, and need a positive account balance. Shared inputs and outputs inform OpenAI's evaluation and training of models.
- Your prompts and data
- Shared traffic informs evaluation and training (source: help.openai.com)
- Beyond the free allowance
- A request that crosses the quota is billed normally (source: help.openai.com)
Terms read on help.openai.com on OpenAI: the free allowance and how to start OpenAI for Startups: more credits for startups
Free credits to start with
A free amount of usage, on signup or every month; after it, usage is paid.
- 1
Alibaba Cloud Model Studio
Trying Qwen models with a token quota per model
New Model Studio accounts in the Singapore region get a separate free quota for each eligible model, typically 1,000,000 tokens, valid for 90 days. It covers real-time inference only, and afterwards usage is billed pay as you go unless Free Quota Only is on.
- Your prompts and data
- Not used to train or improve foundation models (source: alibabacloud.com)
- Beyond the free allowance
- Auto pay-as-you-go unless Free Quota Only is turned on (source: alibabacloud.com)
Terms read on alibabacloud.com on Alibaba Cloud Model Studio: the free allowance and how to start Alibaba Cloud Startup / AI Catalyst Program: more credits for startups - 2
Nebius Token Factory
Trying open models on Nebius with a card on file
New Token Factory accounts get a $1 trial credit valid for 30 days, and billing setup needs a bank card. Nebius's free Builder Program adds $25 of Token Factory credit for learning and testing. Your content is not used to train any models.
- Your prompts and data
- Not used for training; kept for speed unless ZDR (source: docs.tokenfactory.nebius.com)
- Beyond the free allowance
- Free Builder Program adds $25 of Token Factory credit (source: dev.nebius.com)
Terms read on docs.tokenfactory.nebius.com on Nebius Token Factory: the free allowance and how to start Nebius AI Lift for Startups: more credits for startups - 3
Cerebras
Short experiments on gpt-oss-120b or Qwen 3.8 27B
Adding a valid payment method unlocks a one-time $5 credit that expires after 30 days, with Free Trial limits of 5 requests a minute on gpt-oss-120b. It is for development, not production, and access pauses once the credit ends unless you buy credits.
- Your prompts and data
- Not used to train or fine-tune models (source: cerebras.ai)
- Beyond the free allowance
- API access pauses until you buy pay-as-you-go credits (source: cerebras.ai)
Terms read on cerebras.ai on Cerebras: the free allowance and how to start - 4
Fireworks AI
A quick test of open models on Fireworks
Fireworks describes a $1 credit, with account-wide requests capped at 10 a minute until a payment method and credits are on file. Once it is spent the account is suspended until a payment method is added. Fireworks does not log prompts or generations for open models without opt-in.
- Your prompts and data
- Prompts and generations not logged for open models (source: docs.fireworks.ai)
- Beyond the free allowance
- Account suspended until a payment method is added (source: docs.fireworks.ai)
Terms read on docs.fireworks.ai on Fireworks AI: the free allowance and how to start Fireworks for Startups: more credits for startups
Startup programs with more model credits
Programs that give startups credits for these models and the clouds that host them, largest tracked value first. Most ask for an application, and some for a backing investor.
- Catalog valueSource and terms for OpenAI for Startups
How this comparison is built
The list covers model APIs a developer can call for free today without applying to a program, grouped into free tiers with no end date and free credits. The order and the "best for" notes are our editorial reading of each provider's terms. Every term comes from the provider's own pricing, limits or documentation page on the date shown, never from a third-party list; follow each link for the full conditions. Where a provider states how it uses prompts sent on its free allowance, the comparison quotes it, because that matters before you send customer data. The startup programs below are tracked like every program in the catalog, with their own sources and check dates. No provider pays for a placement on this page.
Read the full methodologyCommon questions
Is there a free LLM API?
Yes. Several providers offer free access to their model APIs, either as a free tier with rate limits or as free credits. The comparison on this page lists the ones we track with what each covers. Limits and model lists change often, so every allowance carries the date its terms were read.
Do free LLM APIs need a credit card?
It depends on the provider. Some free tiers work with only an account, while others ask for a payment card or a phone number before any usage. The comparison states what each provider's own page says about it.
Can I use a free AI API tier in production?
Providers describe their free allowances as meant for testing, prototyping or evaluation, and the rate limits are usually too low for production traffic. Some also restrict commercial use or exclude certain regions. Where a provider states such a limit, the comparison quotes it.
Do providers train on prompts sent to a free tier?
Some do, and say so: a free tier can come with terms that let the provider use prompts and outputs to improve its products, while its paid usage does not. Others do not use API data for training at all. The comparison quotes each provider's own statement, which is the one to read before you send anything sensitive.
What do I do when a free allowance runs out?
You move to the provider's paid usage, or you apply to a startup program. Model makers and the clouds that host their models run programs that give startups far larger credits, usually by application. They are listed at the end of this guide, largest first.
More best-of guides
New startup perks, in your inbox
Get an email when a new credit, discount, or deal for founders is added. No spam, unsubscribe anytime.