Alibaba (opens in a new tab)

Qwen4-Max

Qwen4-Max is a model from Alibaba's Qwen team, released on November 16, 2026, which Alibaba calls the most capable model in the Qwen family to date: a 2.4-trillion-parameter Mixture-of-Experts model with native vision and a one-million-token context window. It is offered through QwenCloud's API and Token Plan and supports tool calling and structured outputs.

Released

November 16, 2026

Type

multimodal

License

proprietary

Context

—

Alibaba states a one-million-token context window; the QwenCloud model page lists 131K maximum output for non-thinking responses.

Pricing

Input

$1.50 / 1M tokens

Output

$4.50 / 1M tokens

List price on the developer's own pages, checked . QwenCloud list price (international, ap-southeast-1 endpoint). Implicit-cache input $0.075; explicit cache writes $1.875 (5-minute) and $3 (1-hour). Thinking tokens are billed as output.

Compare API prices

Capabilities

textimagesvideoreasoningtoolscode

Architecture

Parameters: 2.4T total (MoE)

Sources

Every figure on this page was checked against these pages on . A figure we could not source is left out.

Spotted something out of date? Tell us.

Links

In the news: Qwen4-Max and Alibaba

SiliconANGLE AI

Alibaba launches Qwen4-Max with training, attention optimizations

The Decoder

Alibaba's Qwen4-Max takes on Opus 5.5, GPT-6, and Kimi K3.1 Pro at much lower prices

Digital Trends

Qwen4-Max is China’s newest flagship AI model, and it makes GPT and Claude look expensive

InfoWorld AI

Alibaba touts 62% lower task costs with Qwen4-Max, adds cyber safeguards

MarkTechPost

Alibaba Qwen Introduces QwenGyre: Elastic Reinforcement Learning RL for XLong-Horizon Agents

Alibaba's Qwen team has introduced QwenGyre, an elastic reinforcement learning (RL) framework designed for training AI agents on extreme-long-horizon tasks. QwenGyre is an online reinforcement learning system that trains coding agents through unmodified, black-box harnesses, targeting rollouts that can run for hours and process close to 1 million tokens. The framework was tested on Alibaba's Qwen 3.6 122B and Qwen 3.8 2.4T models, with approximately 700,000 tokens per rollout. It operates on multi-node GPU clusters, with the 122B model utilizing 24 to 32 nodes and the 2.4T model using 192 nodes. The system employs two core ideas: it moves idle GPUs between rollout and training, and it transforms branching agent histories into bounded training samples. Performance tests showed QwenGyre to be 1.38x to 1.78x faster than Async and 1.21x to 1.85x faster than Colocate, achieving matched scores. The best performance was a 1.85x end-to-end speedup over Colocate on TerminalBench. QwenGyre addresses challenges in 'XLong-horizon' tasks, defined as agent runs spanning hours with hundreds of model and tool interactions, such as generating full repositories from specifications. Two primary problems at this scale are long rollout tails, where training waits for the slowest execution, and the complexity of branching agent histories, which inflate training costs by duplicating shared prefixes. The research team has not released public code for QwenGyre.

Compare Qwen4-Max with