May 20 News.Ali Chikan (1974-), PRC politician, prime minister 1997-1998, vice-president from 2008Large ModelOfficially released today Qwen 3.7-Max - A new generation of flagship models for the intelligent age, to be served through API. Qwen3.7-Max is the most comprehensive and powerful man ever askedSmart Body Model.
This is the first time that a large-scale model has been introducedQwen3.7-Max is committed to being an all-powerful intellectual base • Be able to prepare and debug codes, automate office processes, and continue to implement autonomously over hundreds or thousands of steps of long-cycle tasks。
The core strength of Qwen3.7-Max lies in the breadth and depth of intelligence capabilities:
- In terms of programming, the development of the front-end prototype to a complex multi-document engineering capability
- WORKSTREAM AUTOMATION THROUGH MCP INTEGRATION AND MULTI-INTELLIGENCE COLLABORATION IN OFFICE AND PRODUCTIVITY
- (a) In the area of long-cycle self-implementation, consistent reasoning was maintained in a 35-hour, over 1,000-time, fully-owned kernel optimization experiment, which validated its capacity for sustained and stable implementation
- In addition, both within Claude Code, OpenClaw and Qwen Code, as well as under other frameworks, they can steadily develop an excellent cross-framework capability。
Qwen3.7-Max is about to provide its services through Alarion. Users can call through API。
1AI with Qwen 3.7-Max Model Test as follows:
In programming intelligenceQwen3.7-Max has led on SWE-Pro (60.6), SWE-Multilingual (78.3), SciCode (53.5) and QwenSVG (1608). Beyond DS-V4-Pro Max (67.9). On SWE-Verified (80.4), it is comparable to Opus-4.6 Max (80.8) and DS-V4-Pro Max (80.6)。
In terms of universal intelligence, higher. Qwen3.7-Max featured prominently on MCP-Mark (57.5 in 60.8 vs. GLM-51), MCP-Atlas (75.8 in 76.4 vs. Opus-4.6), and Skillbench (56.2 vs. K2.6 in 59.2) and demonstrated the ability to optimize the inside of GPU on Kernel Bench L3 (1.98 times the median, 96% acceleration rate). The same performance was shown on BFCL-V4 (75.0), Qwenclaw (64.3) and ClawEval (65.2) in pursuit of Opus-4.6 Max. Score 87.0, top of the office automation benchmark SpreadSheetBench-v1。
In terms of reasoning, Qwen 3.7-Max has advanced in GPQA Diamond (91.3 in 92.4 vs. Opus-4.6), HLE (40.0), HMT 2026 Feb (96.2 in 97.1 vs. Opus-4.6), IIMOnswerBench (90.0 vs. DS-V4-Pro) and Apex (38.3 in 44.5 vs. DS-V4-Pro)。
Universality and multilingualism, Qwen3.7-Max was prominent on IFBench (77.0 in 79.1 vs. DS-V4-Pro) and demonstrated command compliance. The same lead on WMT24++(85.8) and MAXIFE(89.2) indicates that their multilingual understanding and translation quality is at first-class level. Demonstrating on SuperGPQA (73.6) and QwenWorldBench (57.3)。