Qwen3.8-Max Official Release ReportProvider report
2026-08-03Based on the two complete performance comparison tables in Qwen's official August 3, 2026 release article. The source contains 86 benchmark rows; this site splits the paired values for Agents' Last Exam, MathVision, BabyVision, ZeroBench, OSWorld 2.0, ClawEval-MM, CharXiv (RQ), and OCR-Bench-V2 into 94 metrics and 510 scores. The article's AIME24 method-search table evaluates a Qwen3-8B data-selection experiment rather than Qwen3.8-Max itself, so it is excluded. Comparison-model values are preserved as reported, and '--' cells do not create scores.