Arctic puts 1 free language models and 1 free image models from 2 makers behind one key and one API. It runs only on free tiers (Pollinations), so nobody pays for anything.
One prompt, up to four models, streamed side by side with a receipt per lane.
The live catalog: prices, context, tools, reasoning and vision, filterable and exportable.
One visual brief sent to two image models.
Design an agent in conversation, test it, then call it with a key.
OpenAI-compatible Chat Completions and a single-file terminal agent.
Client keys with budgets, rate limits, allowlists, rotation and receipts.
Start in the Playground. Sign a message with your wallet (free, no gas) to get client keys. Point any OpenAI SDK at Arctic and keep your code as it is.
Send model: "auto" and Arctic picks a free model, skipping models that cannot handle tools or images in your request. When a free host is busy, the call moves to another host of the same model.
Open 20/day · Member 200/day · Holder 1,000/day · Guardian 4,000/day requests. Every response names its host; the Ledger shows the whole pool live.
Newest in the catalog: GPT OSS 20B (2025-08-05).