Alibaba released Qwen3.8-Max on Monday morning. The 2.4 trillion-parameter model scored 1,668 points on Arena's Frontend Code leaderboard, landing fourth place.
Only Claude Opus 5 (Max) at 1,705 and Moonshot's Kimi K3 (Max) at 1,676 ranked higher. The model exited its July preview with 95 billion active parameters, a 1 million-token context window, and public API pricing of $2 per million input tokens and $6 per million output tokens.
Qwen also placed second in Arena's Consumer Product category, though Anthropic's Fable 5 still dominates core software engineering benchmarks. On SWE-Pro, Fable scores 80.0 versus Qwen's 67.7, and on FrontierSWE it reaches 88.8 against 73.5.
Alibaba's internal data claims wins over Claude Opus 4.8 and GPT-5.6 on agentic and multimodal tests. TerminalBench-2.1 shows 86.6, PaperBench hits 93.0, and OSWorld-Verified reaches 86.1.
The bigger news: next week, Alibaba will open-source the model weights for the first time at the Max-class tier. Weights land on Hugging Face and ModelScope, alongside an open-weights release of Qwen3.8-27B.
Hong Kong investors moved fast. Alibaba shares climbed 6.15% to HK$124.20 by lunch break, extending Friday's 4.65% rally that analysts tied to a reported Moonshot chip partnership.
Independent evaluations will offer clearer benchmarking once the weights become public. For now, the coding results show Qwen closing ground on Western models while pushing toward developer accessibility through open weights.
This article is for informational purposes only and should not be construed as financial or investment advice.
