Qwen4-Max debuts: 2.4T parameters, 1M context, ~60% cheaper per task

mark_k · x · 2026-11-16

Alibaba's Qwen team released Qwen4-Max, a new flagship model with 2.4 trillion parameters, native vision, and a 1M-token context window. Its MoE architecture activates 71B parameters per token, 25% fewer than its predecessor. Alibaba claims gains in coding, agent tasks and visual reasoning, and says Qwen4-Max used 35% fewer tokens and cut average cost per task by 60% versus Qwen3.8-Max on its own Work360 benchmark. Available via Qwen Cloud and Alibaba apps; smaller open-weight Qwen4 models are coming, with no release date.

Related event: Alibaba Releases Qwen4-Max, Setting New Benchmarks for Chinese Models(55 posts)→

Original post →

More from Models

Models channel →