Table 6; Thinking MoE sampling settings from pages 20-21 Thinking MoE sampling: temperature 0.6, top-p 0.95, top-k 20; presence penalty for MoE is not explicitly stated (the 1.5 clause follows the Dense Thinking settings); max output 32,768 tokens, except AIME-25, HMMT-25 and LiveCodeBench v6 use 81,920. 仅取235B-A22B-Thinking目标列,其他变体不是本结果的替代。
Qwen3-VL-235B-A22B-Thinking · PolyMATH · Qwen3-VL技术报告v2;Table 6; Thinking MoE sampling settings from pages 20-21 · Scores and configurations
Qwen3-VL-235B-A22B-Thinking record for PolyMATH · Qwen3-VL技术报告v2;Table 6; Thinking MoE sampling settings from pages 20-21. 1 score record for this release, including 1 exact value (may include historical records). Includes 1 source location.
Table 6; Thinking MoE sampling settings from pages 20-21 Thinking MoE sampling: temperature 0.6, top-p 0.95, top-k 20; presence penalty for MoE is not explicitly stated (the 1.5 clause follows the Dense Thinking settings); max output 32,768 tokens, except AIME-25, HMMT-25 and LiveCodeBench v6 use 81,920. 仅取235B-A22B-Thinking目标列,其他变体不是本结果的替代。