Free LLM API credits and free access paths

Best for

Learning, small experiments, model comparison, and early prototypes.

Free access note

OpenRouter describes openrouter/free as a router that sends requests to currently available free models, and also supports :free variants for specific models where a free variant exists.

Limits note

Source snapshot, May 2026: OpenRouter docs described free-model limits such as 50 requests per day for accounts that have purchased less than 10 credits, 20 requests per minute, and 1000 free-model requests per day after purchasing at least 10 credits. Check OpenRouter directly before relying on this.

Caution

Free model access can have lower limits, changing availability, provider-side throttling, and temporary unavailability. Do not treat free access as production capacity.

Privacy note

Avoid sensitive, private, or confidential data unless you have reviewed OpenRouter's current terms, data-policy details, and account settings.

Best for

Fast API experiments, latency-sensitive prototypes, testing supported open models, and small developer projects.

Free access note

GroqCloud offers a Free plan for building and testing with its APIs, with paid developer and enterprise options available for higher-scale use.

Limits note

Source snapshot, May 2026: Groq docs describe rate limits using requests per minute, requests per day, tokens per minute, tokens per day, audio seconds per hour, and audio seconds per day. Limits can vary by model, account, organization, and tier, so check GroqCloud's current limits before relying on this.

Caution

Useful for testing and prototypes, but free-plan limits are not production capacity. Applications should handle rate limits and provider-side changes gracefully.

Privacy note

Avoid sensitive, private, or confidential data unless you have reviewed Groq's current terms, data-policy details, and account settings.

Best for

Serverless AI experiments, edge-adjacent prototypes, lightweight inference workflows, embeddings, and developers already using Cloudflare Workers.

Free access note

Cloudflare Workers AI is available on Free and Paid Workers plans, with a daily free allocation for testing and small usage.

Limits note

Source snapshot, May 2026: Cloudflare docs described a free allocation of 10,000 Neurons per day. Usage above the free allocation requires the Workers Paid plan, and operations can fail after limits are exceeded. Check Cloudflare's current docs before relying on this.

Caution

Useful for serverless and edge-adjacent experiments, but the free allocation is not production capacity. Developers should understand Neurons, model availability, quota behavior, and failure handling before relying on it.

Privacy note

Avoid sensitive, private, or confidential data unless you have reviewed Cloudflare's current terms, data-handling details, and account settings.

Best for

Testing Cohere's chat, embedding, rerank, retrieval, and proof-of-concept API workflows.

Free access note

Cohere docs describe evaluation or trial API keys as free but limited, with production keys available for paid and higher-scale use.

Limits note

Source snapshot, May 2026: Cohere docs described trial keys as limited to 1,000 API calls per month, with listed trial Chat API limits such as 20 requests per minute for several chat models. Limits can vary by endpoint, model, key type, and account status, so check Cohere's current docs before relying on this.

Caution

Useful for evaluation and proofs of concept, but trial-key access is limited and should not be treated as production capacity. Developers should review Cohere's current rate limits, pricing, and production-key requirements before relying on it.

Privacy note

Avoid sensitive, private, or confidential data unless you have reviewed Cohere's current terms, data-handling details, and account settings.

Best for

Testing Mistral models, building small API prototypes, trying the Studio playground, experimenting with agents, and evaluating Mistral's chat and developer workflows.

Free access note

Mistral docs describe Free mode as enabling API access by default with no credit card required, while usage and rate limits still apply.

Limits note

Source snapshot, June 2026: Mistral docs described Free mode API access as enabled by default with no credit card required, but with limited usage and rate limits. Check Mistral's current docs before relying on this.

Caution

Useful for testing and early prototypes, but Free mode should not be treated as production capacity. Developers should review current usage limits, API pricing, data terms, and production requirements before relying on it.

Privacy note

Avoid sensitive, private, or confidential data unless you have reviewed Mistral's current terms, data-handling details, and account settings.

Best for

Trying Gemini in a browser, learning the Gemini API, prototyping prompts, and testing Google model features before deciding on a paid path.

Free access note

Checked 11 July 2026: Google said new accounts begin on a Free Tier with access to certain models in AI Studio and the Gemini API, subject to model and account limits. Paid projects have different billing and data-use terms.

Limits note

Google says active rate limits depend on model, project, usage tier, and account status, can update over time, and are not guaranteed. Check the limits shown in AI Studio rather than relying on a saved number.

Caution

Free-tier access is for testing, not a promise of production capacity. Review supported regions, model status, billing setup, spend controls, rate limits, and the terms attached to your project before relying on it.

Privacy note

Google's terms say unpaid-service prompts and responses may be used to improve products and may be reviewed by humans, with a regional exception for the EEA, Switzerland, and UK. Paid-service terms describe a different data-use boundary. Do not send sensitive, confidential, or personal information to unpaid services.

Best for

Trying hosted models across providers, comparing routes, and adding text, image, audio, or embedding tasks without managing inference servers.

Free access note

Checked 11 July 2026: Hugging Face listed monthly Inference Providers credits, including $0.10 for Free users, marked subject to change. Extra usage requires purchased credits; account type affects the allowance.

Limits note

Model availability, provider support, task coverage, pricing, and account credits can change. Check the live model page, routed provider, and current credit balance before each test or deployment decision.

Caution

A common API does not remove provider differences. Confirm task support, model terms, failure handling, prices, and the provider that will run each request before production use.

Privacy note

Hugging Face says it does not store request or response bodies when routing and keeps debugging logs for up to 30 days without user data or tokens. It also directs users to each external provider's data-security policy.

Best for

Testing multi-provider routing, adding fallbacks, and keeping model choices easier to change in Vercel or AI SDK projects.

Free access note

Checked 11 July 2026: Vercel's pricing page showed a $5 monthly AI Gateway credit for team accounts. It said the credit starts with the first request and ends after an account purchases gateway credits. Check the current page before relying on that amount or eligibility.

Limits note

Charges depend on the selected model, provider, and token use. The official page directs readers to the live model list or dashboard for current rates; purchased credits and optional auto top-up can create charges.

Caution

A gateway can simplify switching and fallbacks, but it does not make every provider interchangeable. Review model support, provider terms, budgets, failure behavior, and current prices before production use.

Privacy note

Vercel documents optional routing controls, including zero-data-retention and no-prompt-training filters for supported routes. Availability depends on plan, model, provider, and configuration, so verify the current settings and terms.