Alibaba released Qwen3.8-Max on Monday, a large language model that ranks fourth on Arena's Frontend Code leaderboard. Only Claude Opus 5 and Kimi K3 scored higher. The model has 2.4 trillion total parameters and 95 billion active parameters, with a 1 million-token context window. API pricing is set at $2 per million input tokens and $6 per million output tokens.
How Qwen3.8-Max Stacks Up
On the Arena leaderboard, Qwen3.8-Max scored 1668, behind Claude Opus 5 (High) at 1669, Claude Opus 5 (Max) at 1705, and Kimi K3 (Max) at 1676. In the Consumer Product category, it placed second. Alibaba claims the model outperforms Claude Opus 4.8, Fable 5, and GPT-5.6 on agentic and multimodal benchmarks. Specific scores include 86.6 on TerminalBench-2.1, 93.0 on PaperBench, and 86.1 on OSWorld-Verified.
Open-Source Release and Market Reaction
Qwen3.8-Max is the first open-source Max-class model from Alibaba. The company will release open weights of Qwen3.8-Max and Qwen3.8-27B next week on Hugging Face and ModelScope. Investors responded positively: Alibaba shares climbed 6.15% to HK$124.20 on launch day, extending Friday's 4.65% gain that analysts tied to a reported Moonshot chip deal.
Benchmark Comparisons
While Qwen3.8-Max leads in several areas, Anthropic's Fable 5 still dominates core software engineering benchmarks. Fable 5 scored 80.0 on SWE-Pro against Qwen's 67.7, and 88.8 on FrontierSWE against 73.5. The open weights release next week will let developers test the model's real-world performance for themselves.
