Qwen3.8-Max
Alibaba's largest model yet — a 2.4-trillion-parameter multimodal MoE flagship with 1M context accepting text, image, and video input, and the first Max-class Qwen slated for open weights
Qwen3.8-Max
Qwen • August 2026
Training Data
Not disclosed
Qwen3.8-Max
August 2026
Parameters
2.4 trillion (sparse MoE; ~95B active per press reports)
Training Method
Sparse Mixture of Experts with RL scaling
Context Window
1,000,000 tokens
Knowledge Cutoff
Not disclosed
Key Features
2.4T-Parameter Sparse MoE • Text, Image & Video Input • 1M Context with 131K Output • Open Weights Announced
Capabilities
Multimodal: Outstanding
Coding: Outstanding
Agentic: Outstanding
What's New in This Version
More than doubles Qwen3.7-Max's agentic coding scores (FrontierSWE 40.7 → 73.5, DeepSWE 1.1 21.6 → 56.6), adds image and video input to the Max line, and ranked second in Vision Arena at launch
Alibaba's largest model yet — a 2.4-trillion-parameter multimodal MoE flagship with 1M context accepting text, image, and video input, and the first Max-class Qwen slated for open weights
What's New in This Version
More than doubles Qwen3.7-Max's agentic coding scores (FrontierSWE 40.7 → 73.5, DeepSWE 1.1 21.6 → 56.6), adds image and video input to the Max line, and ranked second in Vision Arena at launch
Technical Specifications
Key Features
Capabilities
Other Qwen Models
Explore more models from Qwen
Qwen3.7-Max
Alibaba's closed text flagship for reasoning and long-horizon agents — 1M context, MCP and multi-agent orchestration, tuned for autonomous coding and office automation
Qwen3.7-Plus
Multimodal agent sibling to Qwen3.7-Max, adding native vision while keeping the 1M-token context and agentic tooling at a lower price point
Qwen3.6-27B
Alibaba's first dense open-weight Qwen3.6 — 27B beats 397B MoE on agentic coding via hybrid Gated DeltaNet + Gated Attention
