Alibaba launches Qwen4-Max: 2.4 trillion parameter model with 1 million token context

Alibaba has announced the general release of Qwen4-Max, the flagship model of the new Qwen4 generation. The model uses a Mixture-of-Experts (MoE) architecture with a 2.4 trillion parameter backbone and activates only 71 billion parameters per token, plus native vision and a context window of up to 1 million tokens.

What happened?

On November 16, 2026, Alibaba made Qwen4-Max generally available via QwenCloud API, Token Plan, Qwen Studio, QwenWork and Qoder. The model accepts text, images and video, and is designed for complex work, creativity and research tasks. The company says it uses up to about 50% fewer tokens than Qwen3.8-Max on comparable tasks and reduces average cost per task by more than 60% on internal benchmarks such as Work360.

Why does this matter?

Qwen4-Max represents an advance in the efficiency of giant models. With fewer active parameters and lower prices (around $1.50 per million input tokens and $4.50 output, with cache at $0.075), it makes frontier AI more accessible for developers and companies. Alibaba plans to release smaller open-weight versions of the Qwen4 series.

What changes in practice?

Developers can already access the model via the QwenCloud API. The combination of long context, native vision and lower cost favors long-running agents, analysis of extensive documents and multimodal workflows. Smaller open-source models are expected soon for those who prefer local runs.

By Grok