Qwen 3.8-Max Gets Further Upgrade: All 8 Programming Test Benchmarks Improve, Some Outperform Opus 5

Beating AI Express: Alibaba’s Qwen has upgraded its flagship model Qwen3.8-Max, which was launched just one month ago, to Qwen3.8-Max-0902. The parameter scale remains unchanged at 2.4 trillion parameters with a 1 million-token context window. This upgrade focuses on post-training for coding and collaboration, with a priority on improving programming, complex workflows, and long-horizon agent tasks. In Alibaba’s published comparison table, the 0902 version outperforms the original in all 8 programming benchmarks. Specific improvements include: TerminalBench 3.0 rising from 11.3 to 29.0, DeepSWE 1.1 from 56.6 to 69.3, QwenSWEbench V2 from 55.1 to 70.0, and professional task benchmark JobBench from 53.4 to 64.0. Compared with Claude Opus 5, Qwen3.8-Max-0902 still lags in most programming projects, but surpasses the competitor in three benchmarks: MLS-Bench-Lite, SWE-Atlas QnA, and QwenSWEbench V2. The four multimodal benchmarks only saw gains of 0.4 to 3 points over the original, making it clear that this upgrade’s core focus is enhancing the model’s coding and task execution capabilities for agents. All the above results are currently Alibaba’s internal tests. Qwen3.8-Max-0902 is now available on the QwenCloud API, with pricing unchanged: $2 per 1 million input tokens and $6 per 1 million output tokens.