Mathematical Reasoning on DAPO-Math OOD average
60.5Avg@16GEOSD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GEOSDBackbone=Qwen3-8B2026.07 | 60.5 | 8.6 | |
| TrOPDBackbone=Qwen3-8B2026.07 | 54.8 | 2.9 | |
| GEOSDBackbone=Olmo-3-7B-Think2026.07 | 54.8 | 7.6 | |
| GRPOBackbone=Qwen3-8B2026.07 | 53.6 | 1.7 | |
| TIPBackbone=Qwen3-8B2026.07 | 53.3 | 1.4 | |
| SkewKLBackbone=Qwen3-8B2026.07 | 52.7 | 0.8 | |
| π0Backbone=Qwen3-8B2026.07 | 51.9 | — | |
| JSDBackbone=Qwen3-8B2026.07 | 51.8 | 0.1 | |
| TrOPDBackbone=Olmo-3-7B-Think2026.07 | 50.4 | 3.2 | |
| TIPBackbone=Olmo-3-7B-Think2026.07 | 50.1 | 2.9 | |
| GEOSDBackbone=DS-R1-Llama-8B2026.07 | 49.4 | 5.7 | |
| SkewKLBackbone=Olmo-3-7B-Think2026.07 | 47.9 | 0.7 | |
| GRPOBackbone=Olmo-3-7B-Think2026.07 | 47.6 | 0.4 | |
| π0Backbone=Olmo-3-7B-Think2026.07 | 47.2 | — | |
| TIPBackbone=DS-R1-Llama-8B2026.07 | 46.8 | 3.1 | |
| GRPOBackbone=DS-R1-Llama-8B2026.07 | 46.2 | 2.5 | |
| JSDBackbone=Olmo-3-7B-Think2026.07 | 46.1 | 1.1 | |
| RevKLBackbone=Qwen3-8B2026.07 | 44.4 | 7.5 | |
| TrOPDBackbone=DS-R1-Llama-8B2026.07 | 44.1 | 0.4 | |
| RevKLBackbone=Olmo-3-7B-Think2026.07 | 43.9 | 3.3 | |
| FwdKLBackbone=Qwen3-8B2026.07 | 43.8 | 8.1 | |
| π0Backbone=DS-R1-Llama-8B2026.07 | 43.7 | — | |
| JSDBackbone=DS-R1-Llama-8B2026.07 | 42.9 | 0.8 | |
| SFTBackbone=Olmo-3-7B-Think2026.07 | 42.8 | 4.4 | |
| FwdKLBackbone=Olmo-3-7B-Think2026.07 | 42.6 | 4.6 | |
| SkewKLBackbone=DS-R1-Llama-8B2026.07 | 41.5 | 2.2 | |
| RevKLBackbone=DS-R1-Llama-8B2026.07 | 40.8 | 2.9 | |
| SFTBackbone=Qwen3-8B2026.07 | 39.4 | 12.5 | |
| SFTBackbone=DS-R1-Llama-8B2026.07 | 38.9 | 4.8 | |
| FwdKLBackbone=DS-R1-Llama-8B2026.07 | 37.8 | 5.9 |