Alibaba Releases Qwen4-Max: 2.4T-Parameter MoE, 71B Active

Alibaba Releases Qwen4-Max: 2.4T-Parameter MoE, 71B Active Params, and 1M-Token Context


The Qwen4-Max release arrives as 2026's frontier landscape shifts decisively toward sparse MoE designs, trillion-scale backbones, and million-token context windows—all now table stakes for flagship enterprise models.


Alibaba's Qwen team has released Qwen4-Max, the first model of its Qwen4 generation. Qwen4-Max is a proprietary sparse MoE model with a 2.4T-parameter backbone and 71B parameters active per token. It accepts text, images, and video, and supports a 1M-token context window. Alibaba also cut API prices and reports a 62% drop in average cost per task.


TL;DR


  • Size: 2.4T backbone parameters, 71B active per token, plus 328B N-gram embedding parameters. 1M context, 393K max output tokens.
  • Runs on: Hosted only—QwenCloud API, Token Plan, Qwen Studio, QwenWork, and Qoder. No open weights, so no self-hosting or quantized builds.
  • Performance: A large jump over Qwen3.8-Max. Competitive with frontier models on agentic work, but behind on hard coding and vision.
  • Best: HLAB-Vals-1.1 (legal agent work): 21.3, versus 11.7 for Kimi K3.1, the next-best model.

What Is Qwen4-Max?


Qwen4-Max is a multimodal MoE language model from Alibaba, built for long-horizon professional work. A Token Plan-only preview opened this month.


How Does the Qwen4-Max Architecture Work?


Content truncated in source. Architecture details were not provided in the excerpt.


Key Takeaways for Developers and Enterprises


  1. MoE efficiency at scale. With only 71B of its 2.4T parameters active per token, Qwen4-Max balances trillion-parameter capacity against practical inference costs—a design pattern now standard among 2026 frontier models.
  2. 1M-token context. The extended context window supports long-document reasoning, large codebases, and extended agentic sessions without chunking.
  3. Hosted-only deployment. The absence of open weights means teams must rely on QwenCloud-hosted APIs; self-hosting and quantized builds are not available.
  4. Cost reduction. The 62% drop in average cost per task, alongside API price cuts, lowers the barrier for production-scale agentic deployments.
  5. Strengths and gaps. Qwen4-Max leads on legal agent benchmarks (HLAB-Vals-1.1) but trails frontier competitors on hard coding and vision tasks.

  6. Summary


    Qwen4-Max marks Alibaba's entry into the 2026 generation of trillion-parameter, million-token-context models. Its sparse MoE design, multimodal input support, and aggressive pricing make it a strong candidate for long-horizon professional and agentic workloads—though teams requiring open weights or best-in-class coding and vision performance may need to look elsewhere.

    via MarkTechPost

Related