Alibaba's proprietary Qwen3.7-Max can run for 35 hours autonomously and supports external harnesses like Anthropic's Claude Code

2026-05-21 · Source: VentureBeat · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Advanced, medium

Summary

Alibaba's Qwen Team has released Qwen3.7-Max, a proprietary AI model designed for autonomous agentic tasks, capable of "~35 hours of continuous autonomous execution." This model demonstrates "long-horizon reasoning" by optimizing an attention kernel on an unfamiliar T-Head ZW-M890 PPU server for 35 hours, achieving a 10.0x geometric mean speedup. It also simulated a startup's one-year lifecycle in YC-Bench, generating \$2.08 million in virtual revenue. Qwen3.7-Max features a 1-million-token context window and 64K maximum output limit, supporting "cross-harness generalization" by integrating with tools like Claude Code via the Anthropic API protocol. Benchmarks show it scored 44.5 on Apex Math Reasoning, surpassing Claude Opus-4.6 Max's 34.5, and 76.4 on MCP-Atlas. Priced at \$2.50 per 1 million input tokens and \$7.50 per 1 million output tokens via Alibaba Cloud, it positions itself as a premium offering, diverging from Alibaba's prior open-source strategy.

Key takeaway

For enterprise architects evaluating advanced agentic AI solutions, Qwen3.7-Max presents a compelling performance-to-cost ratio, significantly undercutting Western frontier models like GPT-5.4 and Claude Opus 4.7 while matching or exceeding their benchmarks. You should consider its 35-hour autonomous execution and cross-harness generalization for complex, long-horizon tasks. However, its proprietary, API-only access from Chinese endpoints necessitates a thorough review of your data sovereignty and compliance requirements before integration.

Key insights

Autonomous AI agents are now capable of sustained, complex task execution over multiple days, marking a new era.

Principles

• Long-horizon reasoning requires environment scaling.
• Cross-harness generalization expands agent utility.
• Proprietary models can lead benchmark performance.

Method

The model was trained with "environment scaling" across diverse agentic environments to maintain coherence and iteratively improve performance over thousands of turns.

In practice

• Plug into existing agent frameworks like Claude Code.
• Automate complex software development tasks.
• Reformat documents using command-line tools.

Topics

• Qwen3.7-Max
• AI Agents
• Autonomous Execution
• Long-Horizon Reasoning
• API Economy
• Model Benchmarking
• Data Sovereignty

Best for: CTO, VP of Engineering/Data, Machine Learning Engineer, AI Engineer, MLOps Engineer, Director of AI/ML

Related on AIssential

• Qwen3.7-Max: Alibaba’s New Agent-First LLM for Coding, Reasoning, and Long-Horizon AI Workflows — Qwen3.7-Max is an agent-first LLM designed for reliable, long-horizon autonomous task execution and complex tool interaction.
• Qwen 3.7 Max: NEW Powerful AI Model! Beats Opus 4.6, Gemini 3.1, Deepseek v4! (Fully Tested) — Qwen 3.7 Max excels in long-horizon autonomous coding and complex task execution, rivaling frontier models in efficiency and performance.
• Alibaba's latest AI model ran autonomously for 35 hours to optimize code for its own custom chip — Alibaba's Qwen3.7-Max demonstrates advanced autonomous agent capabilities, matching top models and performing complex tasks.
• A third era is now emerging, driven by AI agents capable of tackling large tasks independently over extended periods. — The AI landscape is rapidly evolving with the emergence of autonomous cloud AI agents, exemplified by Cursor's internal metrics showing 35% of merged pull requests generated by these agents, shifting…
• Alibaba's Qwen3.7-Plus supports text, video and imagery inputs at low cost of $0.4/$1.6 per 1M token — but it's proprietary — Qwen3.7-Plus offers cost-effective multimodal AI with state preservation, despite its proprietary nature.
• 🇨🇳 Alibaba unveils new Qwen3.5 model for ‘agentic AI era’, Qwen3.5-397B-A17B. Apache 2.0 license — Sparse Mixture-of-Experts and hybrid attention enhance LLM efficiency and long-context performance for agentic AI.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by VentureBeat.