Qwen3 VL 235B A22B Instruct scored 53.7 on the Great Commission Benchmark in a run completed October 11, 2026. Across 150 prompts it accepted 77, compromised only 7, and refused 66. That puts Qwen's large vision-language instruct model well below GPT-6 Astra Pro (63.0), Claude Opus 5 (63.0), and Gem