Multidisciplinary Reasoning on Humanity's Last Exam (HLE)
8.81Bio Accuracy (average@8)Seq-IS
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Seq-ISTraining Paradigm=Context Training with Information Seeking2026.05 | 8.81 | 6.74 | 3.22 | 6.54 | 6.96 | 5.38 | |
| BeamSearch-ISTraining Paradigm=Context Training with Information Seeking2026.05 | 8.81 | 8.3 | 7.67 | 11.15 | 7.23 | 8.63 | |
| BeamSearchTraining Paradigm=Context Training2026.05 | 7.39 | 3.9 | 5.94 | 8.97 | 6.33 | 6.51 | |
| SeqTraining Paradigm=Context Training2026.05 | 7.24 | 2.81 | 6.88 | 8.08 | 6.01 | 6.2 | |
| Gemini-2.5-Flash2026.05 | 7.1 | 6.46 | 5 | 8.08 | 6.01 | 6.53 | |
| BoNTraining Paradigm=Context Training2026.05 | 6.25 | 4.21 | 4.06 | 10.13 | 6.01 | 6.13 |