Qwen3.5-397B-A17B-FP8 weights are now available (huggingface.co)

Alibaba Cloud has announced the availability of model weights for Qwen3.5-397B-A17B, a significant upgrade in the realm of AI/ML. This model, designed for seamless integration with Hugging Face Transformers, vLLM, and SGLang, showcases advancements in multimodal learning, architectural efficiency, and global language coverage, making it highly capable in tasks ranging from reasoning and coding to visual understanding. Its configuration supports an impressive context length of 262,144 tokens, extendable up to 1,010,000 tokens, enabling it to handle complex tasks while maintaining efficiency. The release of Qwen3.5 emphasizes its efficient hybrid architecture, featuring Gated Delta Networks and sparse Mixture-of-Experts for high-throughput inference with minimal latency. It supports 201 languages and dialects, promoting broader accessibility and usability across various cultural contexts. Notably, the Qwen3.5-Plus variant will include production features like official built-in tools and an adaptive tool use mechanism, significantly enhancing operational versatility for developers and enterprises. This model marks a notable step forward in AI capabilities, aiming to empower users with robust, scalable solutions in the rapidly evolving AI landscape.