Mathematical Reasoning on GSM8K (test) (Accuracy, Avg rounds, Avg calls, Wall (s), n)
99AccuracyB1 (3-sample vote)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| B1 (3-sample vote)Protocol=B1, Agents=GPT-5, Claude-Opus-4-6, Gemini-2.5-Pro, Judge=Claude-Opus-4-62026.05 | 99 | 1 | 3 | 10 | 200 | |
| B2 (Fixed-5 baseline)Protocol=B2, Agents=GPT-5, Claude-Opus-4-6, Gemini-2.5-Pro, Judge=Claude-Opus-4-62026.05 | 99 | 5 | 15 | 42.2 | 199 | |
| B3 (Wald-SPRT)Protocol=B3, Agents=GPT-5, Claude-Opus-4-6, Gemini-2.5-Pro, Judge=Claude-Opus-4-6, alpha=0.05, beta=0.05, Rmax=82026.05 | 97 | 1.01 | 4.06 | 12.1 | 200 |