Alibaba launches Qwen4-Max with native vision, 1M context, 60% lower task cost

thoma_gu · x · 2026-11-16

Alibaba's Qwen team released Qwen4-Max, built on a new MoE architecture with native vision and a 1M-token context window.

The team also notes their Video-Index meta-benchmark (115 video benchmarks audited via an "attack pyramid", 840 hardest verified items) was used in Qwen4 evaluation. The paper finds shuffled frames retain a median 96% of full-video accuracy on 51 benchmarks, and with fixed inputs Claude Opus 5 outscores the best open-source model by 37+ points. The leaderboard chart puts an OpenAI model first at 76.7, Gemini 74.8, Claude 73.8, with Qwen and Kimi both at 63.2.

Related event: Alibaba Releases Qwen4-Max with MoE Architecture, Native Vision and 1M Token Context(15 posts)→

Original post →

More from Models

Models channel →