Qwen3.5-397B-A17B

SemiAnalysis's InferenceX page characterizes Qwen3.5-397B-A17B as the first open-weights model in Alibaba's Qwen3.5 series, citing the Alibaba Cloud Qwen team announcement and the Hugging Face model card as primary sources. It reports a February 2026 release window (February 16 via OpenRouter, February 17 via Artificia The same analysis describes Qwen3.5-397B-A17B as an explicitly agentic and multimodal foundation model targeting reasoning, coding, agent capabilities, and multimodal understanding, and notes that early-fusion multimodal training is reported to achieve cross-generational parity with Qwen3 while outperforming the Qwen3-

Alibaba announced on February 16, 2026 that it has open-sourced Qwen3.5, beginning with the Qwen3.5-397B-A17B checkpoint, which Alibaba also refers to as Qwen3.5-Plus. The post describes it as a natively multimodal foundation model trained across text, images, and video, with support extended to 201 languages and diale The same announcement highlights the model's inference-efficiency design, combining a linear attention mechanism with a sparse mixture-of-experts (MoE) architecture to lower compute requirements. According to the post, Qwen3.5-397B-A17B demonstrates strong performance on language understanding and reasoning, code gener

The Baidu Baike entry on Qwen3.5 confirms that Alibaba's Qwen3.5 series, launched on February 16, 2026, comprises two versions — Qwen3.5-Plus and Qwen3.5-397B-A17B — built around a Hybrid Attention Mechanism and a Sparse Mixture-of-Experts architecture optimized for logical reasoning, mathematical computation, and code The same entry traces development milestones in early 2026 — Alibaba's January announcement of the Qwen3.5 plan, a February 9 code merge request appearing on the Hugging Face open-source project page, and showcase events on February 11 — and reports a score of 87.8 on the MMLU-Pro knowledge reasoning benchmark for the