On all 90 AIME 2024–2026 problems, per-request mean accepted length was broad for all three models and accepted length also fluctuated between verification rounds within requests. Reported spans/medians were 8.24–14.96/11.63 for Qwen3-1.7B, 9.24–16.46/12.03 for Qwen3-8B, and 8.05–17.64/12.22 for DS-Llama-8B; Figure 4 additionally prints distribution means of 11.60, 12.02, and 12.25.

On all 90 AIME 2024–2026 problems, per-request mean accepted length was broad for all three models and accepted length also fluctuated between verification rounds within requests. Reported spans/medians were 8.24–14.96/11.63 for Qwen3-1.7B, 9.24–16.46/12.03 for Qwen3-8B, and 8.05–17.64/12.22 for DS-Llama-8B; Figure 4 additionally prints distribution means of 11.60, 12.02, and 12.25. · CiteArk

测量结果

已评估 0/12 项

获得支持 0有争议或冲突 0尚未评估 12

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
model: Qwen3-1.7B	8.24 tokens	—	尚未评估
model: Qwen3-1.7B	14.96 tokens	—	尚未评估
model: Qwen3-1.7B	11.63 tokens	—	尚未评估
model: Qwen3-1.7B	11.6 tokens	—	尚未评估
model: Qwen3-8B	9.24 tokens	—	尚未评估
model: Qwen3-8B	16.46 tokens	—	尚未评估
model: Qwen3-8B	12.03 tokens	—	尚未评估
model: Qwen3-8B	12.02 tokens	—	尚未评估
model: DS-Llama-8B	8.05 tokens	—	尚未评估
model: DS-Llama-8B	17.64 tokens	—	尚未评估
model: DS-Llama-8B	12.22 tokens	—	尚未评估
model: DS-Llama-8B	12.25 tokens	—	尚未评估

报告指标与观测值

Assessment（0）

关联运行（0）

测量结果