Tiers are ours: a model's place within the board counting only the models whose whole confidence interval is above it. S — top 10%, A — up to 30%, B — up to 60%, C — up to 85%, D — the rest.
Quality vs pricerating up, blended price (3 parts input : 1 output) across, USD per 1M tokens
What counts as a free API here
This page lists models you can call through an API without paying, at providers that offer genuine free access: Groq, Google AI Studio, Cerebras, Mistral, Cloudflare Workers AI, Hugging Face and NVIDIA NIM. A model is listed only when one of these providers flags it as free, or its price there is zero.
A zero price that belongs to a subscription or a coding plan is not treated as free. Such offers look free in some price databases, but you pay for them through the plan, so they are left out.
How the list is ordered
Free models are ranked by the text leaderboard: tier and Elo from the LMArena text arena. For each model you see which providers give it away and, when the source publishes them, the rate limits. The ranking method and the tiers are explained on the text leaderboard.
Using a free tier sensibly
Check the limits. Free tiers are capped by requests per minute or per day; some providers raise the cap after a small top-up or once a card is added.
Read the data terms. Some providers may use prompts sent through free access to improve their models.
Plan for change. Free offers are withdrawn or reduced without notice, so do not build a product that depends on one of them alone.
Keep a paid fallback. The same model is often available for a small fee at another provider; the details of each row list them.
Sources and freshness
Offers and limits come from models.dev and LiteLLM and are refreshed daily, like all prices in the section. The date above the table tells you when the data was taken. If a provider page says something different, the provider is right: confirm the conditions there before you rely on them.