Qwen3.8-Max vs Claude Opus 5

This is the clearest current test of whether the Chinese flagship tier has closed the gap or just the price. Qwen3.8-Max, released August 3, 2026, is a 2.4 trillion parameter Mixture-of-Experts activating roughly 95 billion per token, priced at $2 per million input and $6 per million output, with a 1 million token context window and native text, image, and video input. Claude Opus 5, released July 24, costs $5/$25, ships the same 1 million token window with 128K output, and takes text and vision only. On raw sticker price Qwen is 60 percent cheaper on input and 76 percent cheaper on output, and its cache rates ($0.25 implicit reads, $0.17 explicit reads per million) go further than anything Anthropic publishes. The benchmark picture does not resolve cleanly in either direction. Qwen reports 86.6 on Terminal-Bench 2.1, ahead of Claude Opus 4.8 and Fable 5 at 84.6, and leads PaperBench at 93.0 and IFBench at 82.8. On core software engineering it is well behind: 67.7 against Fable 5's 80.0 on SWE-bench Pro and 73.5 against 88.8 on FrontierSWE, and Opus 5 outscores Fable 5 on most of Anthropic's published rows. Both sets of numbers are vendor-reported. The practical split: Qwen3.8-Max is the value pick for agentic tool use, instruction following, and multimodal work including video, while Opus 5 remains the pick where repository-scale code correctness and long-horizon reliability decide the outcome. Data residency and procurement policy will settle it for many teams before any benchmark does.