AsianFin -- Alibaba’s large language model unit Tongyi Qianwen has released an updated version of its flagship Qwen3 series, boosting long-text processing capabilities to 256K tokens.

The latest model, named Qwen3-235B-A22B-Instruct-2507-FP8, is an upgrade from the previous Qwen3-235B-A22B-FP8 (Non-thinking) version. It has been designed for better instruction-following performance while maintaining efficient computation in FP8 precision.

The new model is now available as open source on both ModelScope and Hugging Face, marking another step in Alibaba’s effort to expand the reach of its open large model ecosystem.