Qwen3.8-Max-0902

A September 2, 2026 post-training snapshot of Qwen3.8-Max (alias `qwen3.8-max-2026-09-02`), not a new model underneath — Alibaba's own changelog describes deeper coding on engineering-scale projects, stronger multi-tool orchestration and refined chart/document/multimodal perception, with the 1M context, thinking mode and tool ecosystem unchanged. Pricing is unchanged at $2/$6 per Mtok ($0.25 cached). The bare `qwen3.8-max` endpoint transitions to this snapshot on September 5, 2026 at the same rates. Only three rows are recorded: Alibaba published a release chart rather than a full comparison table, and its own framing is that the strongest evidence is the snapshot-over-snapshot improvement rather than a clean cross-lab ranking, since the comparison models are run under different harnesses. Recorded values come from BenchmarkList's transcription of that chart, which links the QwenCloud model page and the Alibaba_Qwen launch post as its sources. READ ITS RANK WITH THAT IN MIND: three rows across two categories is exactly the D20 coverage floor, and all three are vendor-chosen rows on which the snapshot does well, so its overall position rests on far thinner evidence than the models around it. It should move once a fuller comparison table or an independent evaluation lands.