Reasoning on GPQA Diamond (accuracy, delta, error correction)
61.11AccuracySST
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SSTcompute=staged compute, imax=4, decoding=greedy argmax2026.04 | 61.11 | 15.15 | 28.04 | |
| Matched fine-tuned baselinecompute=one forward pass per token, decoding=greedy argmax2026.04 | 45.96 | — | — |