Scientific Reasoning on SciKnowEval L3
81Chemistry ScoreDistIL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DistILBackbone=Olmo3-7B-Instruct, Training Time=5h2026.06 | 81 | 74.5 | 55.3 | 76.9 | 71.9 | |
| DistILBackbone=Qwen3-8B, Training Time=5h2026.06 | 80.8 | 80.8 | 66.6 | 76.2 | 76.1 | |
| SDPOBackbone=Qwen3-8B, Training Time=5h2026.06 | 80.2 | 72.4 | 63.6 | 75.9 | 73 | |
| SDPOBackbone=Olmo3-7B-Instruct, Training Time=5h2026.06 | 79.2 | 64.9 | 52.9 | 78.1 | 68.8 | |
| DistILBackbone=Qwen3-8B, Training Time=1h2026.06 | 75.8 | 72.7 | 53.3 | 74.9 | 69.2 | |
| GRPO (off-policy)Backbone=Qwen3-8B, Training Time=5h2026.06 | 74.5 | 72.7 | 59.9 | 77.1 | 71.1 | |
| SDPOBackbone=Qwen3-8B, Training Time=1h2026.06 | 73 | 68 | 52.9 | 72.2 | 66.5 | |
| DistILBackbone=Olmo3-7B-Instruct, Training Time=1h2026.06 | 72.1 | 67.4 | 47.8 | 73.5 | 65.2 | |
| SDPOBackbone=Olmo3-7B-Instruct, Training Time=1h2026.06 | 70.2 | 59.8 | 49.5 | 71.8 | 62.8 | |
| GRPO (off-policy)Backbone=Qwen3-8B, Training Time=1h2026.06 | 65.9 | 63.8 | 35.1 | 74.3 | 59.8 | |
| GRPO (on-policy)Backbone=Qwen3-8B, Training Time=5h2026.06 | 63.4 | 63.6 | 49.8 | 74.1 | 62.7 | |
| GRPO (on-policy)Backbone=Qwen3-8B, Training Time=1h2026.06 | 63.3 | 63.6 | 49.8 | 73.9 | 62.7 | |
| GRPO (on-policy)Backbone=Olmo3-7B-Instruct, Training Time=5h2026.06 | 57.5 | 62.7 | 49.8 | 73.5 | 60.9 | |
| GRPO (off-policy)Backbone=Olmo3-7B-Instruct, Training Time=5h2026.06 | 56.7 | 63.3 | 55.8 | 75 | 62.7 | |
| GRPO (on-policy)Backbone=Olmo3-7B-Instruct, Training Time=1h2026.06 | 51.4 | 62.7 | 49.8 | 73.3 | 59.3 | |
| Qwen3-8BBackbone=Qwen3-8B2026.06 | 41.2 | 59.2 | 30.8 | 58.9 | 47.5 | |
| GRPO (off-policy)Backbone=Olmo3-7B-Instruct, Training Time=1h2026.06 | 39.7 | 55.3 | 35.6 | 70.9 | 50.4 | |
| Olmo3-7B-InstructBackbone=Olmo3-7B-Instruct2026.06 | 22.8 | 37.7 | 16.2 | 36.7 | 28.4 |