Bilibili Open-Sources Index-Translate, a Qwen3.5-Based Translation Model Family for 150 Languages
Bilibili's Index LLM team released Index-Translate under Apache-2.0: 2B, 9B and 35B-A3B preview text models for 150 languages, plus dubbing, syllable-control and long-document variants.
Bilibili's Index LLM team has open-sourced Index-Translate, a family of multilingual translation models based on Alibaba's Qwen3.5, supporting 150 languages and offering specialized variants for dubbing, syllable control, and long-document translation.
What happened
Bilibili's Index LLM team released Index-Translate, a multilingual translation model family based on Qwen3.5, supporting 150 languages. The models are available in 2B, 9B, and 35B-A3B sizes, with specialized variants for dubbing, syllable control, and long-document translation. The models are open-sourced under Apache-2.0 and include benchmarks, code, and an online demo.
Why it matters
Index-Translate is significant for AI developers and NLP researchers due to its multilingual capabilities and specialized variants. It offers practical tools for content creators and businesses dealing with multilingual content, enhancing translation accuracy and consistency.
- Index-Translate offers a comprehensive multilingual translation solution with specialized variants for specific use cases.
- The models are open-sourced under Apache-2.0, making them accessible for developers and researchers.
- Bilibili's benchmarks highlight the models' performance in instruction following and low-resource scenarios.
- The release includes tools like a browser extension and a video dubbing pipeline for practical use.
The bigger picture
Index-Translate has the potential to improve multilingual content creation and translation workflows, particularly for dubbing, subtitles, and long-document translation. Its open-source nature encourages further development and integration into various applications.
Bilibili's Index-Translate represents a significant advancement in multilingual translation, offering specialized models for various applications and open-sourcing them for broader use and development.
Technical details
- Models are built on Alibaba's Qwen3.5 and available in 2B, 9B, and 35B-A3B sizes.
- Specialized variants include Index-Echo (speech-to-speech), Index-Homura (syllable control), and Index-NativeLong (long-document translation).
- The models maintain structure, terminology, and user-specified content during translation.
- Bilibili's benchmarks include FLORES (COMET-22), WMT26 judge scores, and instruction-following tests.
FAQ
What is Index-Translate?
Index-Translate is a family of multilingual translation models built on Alibaba's Qwen3.5, supporting 150 languages and offering specialized variants for dubbing, syllable control, and long-document translation.
What sizes are available for the models?
The models are available in 2B, 9B, and 35B-A3B sizes, with the 35B-A3B version being a preview.
What are the specialized variants of Index-Translate?
The specialized variants include Index-Echo for dubbed speech, Index-Homura for syllable control, and Index-NativeLong for long-document translation.
What benchmarks were used to evaluate the models?
The models were evaluated on benchmarks such as FLORES (COMET-22), WMT26 judge scores, and instruction-following tests.
What tools are included with the release?
The release includes a browser extension for translating web pages and a video dubbing pipeline.
This summary and analysis were generated with AI from the original article and may contain errors. Read the original for the full story.
Read the full story
Originally published by Pandaily on Oct 1, 2026.
Comments
Sign in to join the discussion.