Free tokens: two ways to get text model APIs for nothing

For text models we currently recommend AMD Radeon Cloud — AMD’s official free model API. Free models from other vendors are supported too (paid ones work as well, but there is no need for them).

• ✓How to get a key: sign in to Token Factory and open Public Free Model APIs (phone, email or GitHub login is enough — no credit card needed).
• ✓Pick the OpenAI-compatible protocol and use the base URL below.
• ✓The free shared endpoint offers several mainstream models (e.g. DeepSeek-V4-Flash, Qwen3.8-Flash-Next, MiniCPM5-1B); one key works for all of them — check the actual list on the page.
• ✓Note: the shared endpoint is rate-limited per key and per IP with a daily cap; it suits debugging and light text tasks, not a production-grade high-concurrency API.

OpenRouter aggregates models from many vendors behind one OpenAI-compatible endpoint, and a large part of them come with a free variant: if the model ID ends in `:free` you can call it for free, and one key works for every model. Handy when you want many models at once or want to compare their writing styles.

• ✓How to get a key: sign up at OpenRouter (Google or GitHub login works), then create an API key on the Keys page — no credit card needed.
• ✓Example free models: `z-ai/glm-5.2:free`, `qwen/qwen3.8-27b:free`, `google/gemma-4-31b-it:free`, `nvidia/nemotron-3-super-120b-a12b:free` and more; the list changes over time, so check the official free model list.
• ✓Rate limits: free variants have per-minute and per-day request caps, tiered by the credits your account has purchased historically, and extra accounts or keys do not bypass them; an upstream provider can also return 429, in which case retrying is enough.
• ✓How to use it: the protocol is OpenAI-compatible again, the base URL is below, and you just put the model id (with the `:free` suffix) into the provider’s model list.