Alibaba's Qwen 3.7 Max tops coding benchmarks, T-Head unveils M890 chip
Qwen 3.7 Max beats Claude Opus 4.6 and DeepSeek V4 Pro on agentic coding tests
Alibaba has announced its proprietary Qwen 3.7 Max model achieved leading scores on agentic coding benchmarks, specifically 69.7% on Terminal-Bench 2.0 and 60.6% on SWE-Bench Pro. These scores outperform competitors such as Claude Opus 4.6 and DeepSeek V4 Pro in certain agentic tasks, marking a significant milestone for Alibaba's AI capabilities in code generation and autonomous software development.
In parallel, Alibaba's semiconductor arm T-Head unveiled the Zhenwu M890, described as the company's most advanced AI accelerator chip to date. The chip is designed for both training and inference, positioning Alibaba to reduce reliance on external suppliers and strengthen its full-stack AI infrastructure from model to silicon.
- Qwen 3.7 Max scored 69.7% on Terminal-Bench 2.0 and 60.6% on SWE-Bench Pro, beating Claude Opus 4.6 and DeepSeek V4 Pro in agentic coding tests
- Alibaba's T-Head subsidiary launched the Zhenwu M890 AI accelerator chip, designed for both training and inference workloads
- These releases signal Alibaba's deepening vertical integration from AI models to custom hardware
Why It Matters
Alibaba is closing the gap with Western AI leaders in both model performance and custom silicon for AI workloads.