88Z AI, 6 October — Bilibili's Index LLM team has released the Index-Translate multilingual translation model family, featuring preview weights for 2B, 9B, and 35B-A3B parameter sizes. Built on the Qwen3.5 architecture, the models support translation across 150 languages and have been open-sourced on Hugging Face and ModelScope. Beyond basic instruction following for terminology and format retention, the models are optimized for speech subtitle and dubbing scenarios, allowing translation length adjustment by target
At a glance
- Index-Translate is built on Qwen3.5, supports 150 languages, and has open-sourced 2B/9B/35B-A3B versions.
- The model supports terminology control, syllable-based length adjustment, and consistent long-document translation, suitable for subtitle and dubbing scenarios.
The update
The Index LLM team at Bilibili has officially launched the Index-Translate multilingual translation model family. Preview weights for the 2B, 9B, and 35B-A3B versions are now open-sourced on both Hugging Face and ModelScope platforms. Built upon the Qwen3.5 underlying architecture, this model family covers 150 languages worldwide, catering to diverse cross-language communication needs. In terms of functionality, Index-Translate allows users to precisely control terminology translation, format preservation, and content retention through specific instructions. Furthermore, the model expands into several advanced application capabilities, including optimizations for speech subtitle and dubbing scenarios, dynamic adjustment of translation length based on target language syllable counts, and ensuring translation consistency within the context of long documents, providing robust technical support for professional localization tasks.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- IT之家Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.