Qwen4 Max
Qwen4 Max is the flagship multimodal model in Alibaba's Qwen4 series, succeeding Qwen3.8-Max. Built as a sparse mixture-of-experts (MoE) architecture, it features a 2.4-trillion-parameter total backbone while activating approximately 71 billion parameters per token during inference. The model accepts text, image, and video inputs and provides native deep reasoning capabilities designed for complex software development, office productivity, and agentic workflows.
Architecture and Capabilities
The model features a 1-million-token context window supporting long-horizon professional tasks and extensive code repository analysis. Qwen4 Max includes native reasoning behavior enabled by default, with adjustable effort tiers (such as low, high, and xhigh) to govern chain-of-thought processing. It supports parallel tool calling, web search integration, and structured outputs conforming to JSON schemas or objects. Unlike several previous iterations in the Qwen family, Qwen4 Max is offered as a closed-weight cloud model via managed APIs rather than an open-weight download.
Create with Crafiq
Generate images, 3D models, video and audio in one studio.
Explore the studioHow Qwen4 Max ranks
Qwen4 Max is highlighted in the table below. Switch the metric to see how the ordering changes.