Qwen3 Technical Report (arXiv:2505.09388), Table 11 — Qwen3-235B-A22B (Thinking)

Read https://arxiv.org/html/2505.09388v1 for the benchmark figures on models/qwen3-235b-a22b. As recorded by the reader: Values are official published scores for Qwen3-235B-A22B in thinking mode (Table 11 of the Qwen3 Technical Report, which the HF model card links to). IFEval is the report's 'IFEval strict prompt' row. Non-thinking mode (Table 12) is much lower (GPQA Diamond 62.9, MATH-500 91.2, AIME'24 40.1, AIME'25 24.7, Arena-Hard 96.1) and is not reported here. The HF model card itself contains no benchmark table (only a 235B/22B spec sheet and links), and the Qwen3 blog shows its numbers only as images. No MMLU, MMLU-Pro, SWE-bench Verified, LiveCodeBench v6, or Terminal-Bench 2.1 was published for this model in the official tables (its LiveCodeBench row is v5, 70.7, excluded).