Which AI providers give you free API access
A one-time check of all 10 providers we track — not part of the regular price-verification cycle, since free-tier terms move on each provider's own schedule.
Every price on PerTokens is the standard paid rate from a provider's own pricing page — the number you're actually billed. This page is different: it's a separate check of whether each provider also offers some form of free access before you pay anything. "Free" means very different things provider to provider — an ongoing rate-limited tier, a time-boxed trial credit, or nothing at all — so each row below is checked against that provider's own docs rather than assumed. Once you've picked a free option to test with, compare it against the cheapest models to see what you'd actually pay once you're past it.
| Provider | Free access | What you get | Source |
|---|---|---|---|
| Yes | Gemini API gives ongoing free access to its Flash-tier models with published rate limits — but free-tier requests may be used to improve Google's products, unlike the paid tier. | ai.google.dev — Gemini API pricing ↗ | |
| Alibaba | Trial only | New Model Studio accounts get 1 million free tokens on Qwen-Max and Qwen-Plus tier models, valid for 90 days — Singapore region only, no free quota elsewhere. | alibabacloud.com — Model Studio pricing ↗ |
| Cohere | Trial only | A free evaluation API key covers Command A at 20 requests per minute, capped at 1,000 calls a month — enough to test, not to run in production. | docs.cohere.com — Rate limits ↗ |
| Anthropic | Trial only | New developer accounts get a small amount of free credit to test the API; Anthropic doesn't publish the exact amount, and there's no ongoing free tier beyond it. | platform.claude.com/docs — Pricing ↗ |
| OpenAI | No | New API accounts are enrolled in prepaid billing and must add funds before making a request — there's no free trial credit. | help.openai.com — Prepaid billing ↗ |
| Mistral | No | No free tier for the models we track. A separate experimental endpoint is offered free for feedback, but not the production models listed on this site. | mistral.ai — API pricing ↗ |
| xAI | No | No free tier or trial credit is documented on xAI's own pricing page. | docs.x.ai — Pricing ↗ |
| DeepSeek | No | No free tier documented — usage is deducted from a topped-up (or provider-granted) balance from the first request. | api-docs.deepseek.com — Pricing ↗ |
| Amazon | No | Amazon Bedrock has no dedicated free tier for foundation-model token usage; the general AWS Free Tier doesn't cover per-token inference. | aws.amazon.com — Bedrock pricing ↗ |
| Meta | No | No free tier or trial credit is documented for Llama models on Together AI's pricing page. | together.ai — Pricing ↗ |
This page is checked separately from the 48-hour price-verification cycle that covers the rest of PerTokens — see the date above for when it was last confirmed against each provider's own docs.
Frequently asked questions
No. Every other page on PerTokens shows each model's standard paid per-token rate — what you're billed once you're past any free allowance. This page is the only place that covers free access separately.
Free-tier terms are promotional and change on each provider's own schedule, not ours. We checked each one against its official docs on the date shown above — verify directly before relying on one for anything beyond a first test.
It depends on the provider. Google states that Gemini API free-tier requests may be used to improve its products, while its paid tier is not — a real tradeoff worth knowing before you send anything sensitive through a free key. Check the linked source for any provider before assuming either way.
Google's Gemini free tier comes closest — it's ongoing and rate-limited rather than a one-time credit. The others we found are trial-scoped: time-boxed, region-locked, or capped low enough that they're really only good for a first look.