Qwen3.7-Max - A new generation flagship large model launched by Ali Tongyi

What is Qwen3.7-Max?

Qwen3.7-Max is a new generation flagship model launched by Alibaba's Tongyi Qianwen team, designed for the era of intelligent agents and positioned as an all-around intelligent agent foundation. The model possesses four core capabilities: cutting-edge programming, office automation, long-cycle autonomous execution, and cross-framework generalization. It has achieved leading results on dozens of programming, intelligent agent, and inference benchmarks such as SWE-Pro, MCP-Atlas, and GPQA Diamond, and can be seamlessly integrated into mainstream intelligent agent frameworks such as Claude Code, OpenClaw, and Qwen Code.

Main functions of Qwen3.7-Max

• Cutting-edge programming intelligent agentsIt supports full-chain code writing and debugging from front-end prototyping to complex multi-file software engineering, and performs well on programming benchmarks such as SWE-Pro and SWE-Multilingual.
• Office Productivity AssistantIt achieves workflow automation through MCP integration and multi-agent collaboration, scoring 87.0 on the SpreadSheetBench-v1 office automation benchmark, and can handle complex data analysis and document generation tasks.
• Long-term autonomous executionIt possesses the ability to continuously and stably execute ultra-long-duration tasks, and has maintained coherent inference in a fully autonomous kernel optimization experiment lasting up to 35 hours and involving more than 1,000 tool calls.
• Cross-framework generalizationIt natively adapts to mainstream intelligent agent frameworks such as Claude Code, OpenClaw, and Qwen Code, and can perform stably without the need for fine-tuning for specific frameworks.

Technical Principles of Qwen3.7-Max

• Environmental extension trainingBased on the Qwen3.5 environment extension method, the quality and diversity of the agent training environment are greatly expanded, enabling the model to generalize from diverse environments.
• Decoupled Rollout InfrastructureThe training instance is decoupled into three orthogonal components: task, running framework, and validator, which supports reinforcement learning training across frameworks and validators, forcing the model to learn generalized problem-solving strategies.
• Combinatorial expansionThe same task can be freely recombined with different types and versions of frameworks and validators at extremely low marginal cost, enabling combinatorial scalability of the training environment.
• Long-range reinforcement learning optimizationThrough continuous feedback iteration during long-term autonomous execution, the model can still find substantial improvements after more than 30 hours, verifying its long-term optimization and self-evolution capabilities.

How to use Qwen 3.7-Max

The Qwen3.7-Max plan will be implemented through...Alibaba Cloud Hundred RefinementsProvide services.

Qwen3.7-Max's core advantages

• Leading in all benchmarks for intelligent agentsIt surpasses or closely follows Claude Opus-4.6 Max on general intelligent agent benchmarks such as MCP-Mark, MCP-Atlas, ClawEval, and QwenClawBench.
• Top-notch programming skillsSWE-Pro 60.6, SWE-Multilingual 78.3, Terminal Bench 2.0 69.7: comprehensively leading similar models.
• Profound reasoning and knowledgeWith scores of GPQA Diamond 92.4, HMMT 2026 Feb 97.1, and HLE 41.4, it is among the top tier in high-difficulty STEM reasoning.
• First-class multilingual abilityWMT24++ 85.8, MAXIFE 89.2, MMLU-Pro 89.6: top-tier translation and cross-language understanding quality.
• Real Productivity Closed LoopIt can compress complex projects that would normally require one to two weeks of dedicated team work into end-to-end delivery within hours.
• Hardware-independent generalizationOn the PingtouGe Zhenwu M890 hardware platform, which was not seen during training, it was able to complete deep kernel optimization through independent exploration.

Comparison of Qwen3.7-Max with similar competing products

Application Scenarios of Qwen3.7-Max

• Complex software developmentAs an AI software engineer, I independently completed the entire development lifecycle, including requirements analysis, architecture design, multi-file coding, debugging, and performance optimization.
• Enterprise workflow automationMCP connects to the enterprise toolchain, automatically performing high-intensity office tasks such as data analysis, report generation, and cross-system information integration.
• Underlying system optimizationOn unfamiliar hardware platforms, we independently write, compile, analyze, and iteratively optimize GPU kernels to achieve orders-of-magnitude acceleration.
• Scientific Research and Mathematical ReasoningIt undertakes tasks involving highly complex mathematical proofs, scientific computing, and literature integration, assisting researchers in handling challenging reasoning tasks.
• Multilingual content productionLeveraging top-tier multilingual capabilities, we achieve high-precision translation, cross-language technical document writing, and global content adaptation.