Bilibili opens a free Index-Translate API with no key needed for 150 languages

Bilibili's Index LLM team has open-sourced Index-Translate, a translation model family built on Qwen3.5. The text models cover 150 languages and come in 2B, 9B, and 35B-A3B preview sizes. The team now runs the 35B-A3B model as a free public API, so you can call it without downloading weights or setting up a GPU.

The family also includes three spin-off models. Index-Echo turns speech into translated subtitles, or into translated speech that keeps the original speaker's voice. Index-Homura translates toward a target syllable count, which helps with dubbing and other timed work. Index-NativeLong translates whole documents and keeps the wording consistent across passages.

What you get

The free API serves Index-Translate-35B-A3B at https://index-translate.bilibili.com/v1. It works with both the OpenAI Chat Completions and Responses formats, and requests do not need an API key. No usage cap or end date has been announced for the free access.

Beyond translation quality, the model is trained to follow instructions. You can pass a glossary, set a tone, or ask it to keep JSON or Markdown formatting intact. For fiction, you can pin character names and terms, or tell it to return only the translation with no commentary.

How to use

1. In any OpenAI-compatible client, set the base URL to https://index-translate.bilibili.com/v1 and the model to Index-Translate-35B-A3B. If the client insists on a key, any placeholder string works.
2. Or grab the official call_api.py script, which has no third-party dependencies. Run python call_api.py "你好，世界" -t en to translate Chinese into English. Add --serve to start a local OpenAI-compatible endpoint that browser translation extensions such as Immersive Translate can use.
3. To see the quality first, try the Index-Translate online demo.

Details: Index-Translate on GitHub and the call_api.py script. Weights are on Hugging Face and ModelScope.