Alibaba’s Qwen3.7-Max, a proprietary AI model, can run for 35 hours autonomously and supports external harnesses like Anthropic’s Claude Code. This development marks a significant milestone in the AI industry, which has entered the “agent era” where AI models can plan, execute, and course-correct complex tasks over days, leveraging efficient memory usage.
The Qwen Team, a group of AI researchers at Alibaba, designed Qwen3.7-Max to overcome the limitations of traditional language models, which often degrade when forced to maintain a single train of thought over thousands of conversational turns.
Qwen3.7-Max was trained across a vast array of dynamic agentic environments, allowing it to simulate a one-year lifecycle of a startup and navigate hundreds of decision-making rounds.
In a demonstration of its capabilities, Qwen3.7-Max optimized an attention kernel on an isolated server equipped with a T-Head ZW-M890 PPU, achieving a 10.0x geometric mean speedup over 35 hours.
By comparison, Chinese competitor models like z.ai’s GLM-5.1 and Moonshot’s Kimi K2.6 capped out at 7.3x and 5.0x speedups, respectively.
The model’s endurance is achieved through “environment scaling,” which allows it to simulate complex scenarios and make decisions based on a vast array of data, making it a key component in the development of search box capabilities.
Qwen3.7-Max has built-in reward-hacking self-monitoring, autonomously detecting when it attempts to cheat a training environment and adding heuristic rules to correct its own behavior.
From a product perspective, Qwen3.7-Max is designed to be the cognitive engine for modern software development and enterprise automation, offering a 1-million-token context window and a 64K maximum output limit, which can be particularly useful for companies looking to leverage the power of personalization.
One of its most compelling features is “cross-harness generalization,” which allows it to act as a drop-in intelligence layer for diverse agent frameworks, including support for the Anthropic API protocol.
The benchmark data provided by Alibaba indicates that Qwen3.7-Max has paid massive dividends, scoring 44.5 on the Apex Math Reasoning benchmark and 76.4 on the realistic coding agent benchmark MCP-Atlas.
Developers accessing the API via Alibaba Cloud Model Studio will pay $2.50 per 1 million input tokens and $7.50 per 1 million output tokens.
Qwen3.7-Max occupies a strategic middle ground in the current API economy, demanding a notable premium over domestic rivals but drastically undercutting Western frontier giants, as reported by the outlet.
For context, running heavy agentic workflows through OpenAI’s GPT-5.4 or Anthropic’s Claude Opus 4.7 will run developers $17.50 and $30.00 per million tokens, respectively.
The reaction from the developer community has been mixed, with many praising the engineering achievement but expressing frustration over the licensing model, which locks Qwen3.7-Max behind an API, according to journalists on the scene.
Prominent AI commentator Sudo su (@sudoingX) captured the prevailing sentiment, stating that “qwen is unreal” and that the model’s endurance is “the agent era thing actually happening, not a slide,” as noted by reporters.
However, the loss of the model weights is seen as a blow to the localized AI movement, which relies on state-of-the-art open models to push the boundaries of what can be done on consumer hardware or private enterprise clusters, a point also made by the report.
As Sudo su observed, “one thing though, please open source this one too,” highlighting the importance of accessibility and democratization in the AI community, as emphasized by the filing.
Qwen3.7-Max proves that the autonomous agent era is no longer a theoretical projection; it is a present reality capable of executing complex engineering feats while humans sleep, a fact confirmed by the outlet.
Leave a Reply