Qwen3.7-Max Officially Released: 1,158 self-written code sessions in 35 hours, producing a 10x accelerated operator on domestic chips

robot
Abstract generation in progress
CoinWorld News, Qwen 3.7-Max officially released, after 35 hours of autonomous coding 1,158 times, achieved 10x acceleration on the domestically produced Hengxiao Zhiwu M890 processor. Official data shows that the new model completed a fully autonomous core optimization task without chip architecture documentation or performance analysis data, significantly improving Triton operator performance. The model went through five core evolutionary stages, ultimately achieving a key architecture-level specialized reconstruction. Optimization tests show that Qwen 3.7-Max achieved a 10.0x geometric mean acceleration ratio, surpassing glm 5.1 (7.3x) and kimi k2.6 (5.0x). On general intelligence benchmarks MCP-MARK (60.8 points) and SpreadsheetBench (87.0 points), Qwen 3.7-Max demonstrated strong generalization ability, with overall performance now close to Claude-4.6-Opus-Max.
View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pinned