Multi-step reasoning on GAIA (test)
31Level 1 Accuracy
Skill-R1 (GRPO, Qwen3-4b)
Updated 22d ago
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
2026.05 | 31 | 28 | 10 | 69 | |
| 22 | 21 | 8 | 51 | ||
2026.05 | 21 | 24 | 4 | 49 | |
2026.05 | 4 | 6 | 0 | 10 |
Skill-R1 (GRPO, Qwen3-4b)
| Method | Links | ||||
|---|---|---|---|---|---|
2026.05 | 31 | 28 | 10 | 69 | |
| 22 | 21 | 8 | 51 | ||
2026.05 | 21 | 24 | 4 | 49 | |
2026.05 | 4 | 6 | 0 | 10 |