Mathematical Reasoning Accuracy (%) on OlympiadBench
56.7AccuracyFlowTracer
Evaluation Results
| Method | Links | |
|---|---|---|
| FlowTracerBackbone=Qwen3-8B-Base, Context Length=8K2026.06 | 56.7 | |
| AttentionBackbone=Qwen3-8B-Base, Context Length=8K2026.06 | 55.5 | |
| RandomBackbone=Qwen3-8B-Base, Context Length=8K2026.06 | 55.3 | |
| GRPOBackbone=Qwen3-8B-Base, Context Length=8K2026.06 | 54.4 | |
| GradientBackbone=Qwen3-8B-Base, Context Length=8K2026.06 | 53.4 | |
| CorrelationBackbone=Qwen3-8B-Base, Context Length=8K2026.06 | 53.2 | |
| High-entropyBackbone=Qwen3-8B-Base, Context Length=8K2026.06 | 53.1 | |
| FlowTracerBackbone=Qwen3-4B-Base, Context Length=8K2026.06 | 53 | |
| AttentionBackbone=Qwen3-4B-Base, Context Length=8K2026.06 | 51.5 | |
| GradientBackbone=Qwen3-4B-Base, Context Length=8K2026.06 | 51.3 | |
| RandomBackbone=Qwen3-4B-Base, Context Length=8K2026.06 | 50 | |
| GRPOBackbone=Qwen3-4B-Base, Context Length=8K2026.06 | 49.9 | |
| High-entropyBackbone=Qwen3-4B-Base, Context Length=8K2026.06 | 48.6 | |
| CorrelationBackbone=Qwen3-4B-Base, Context Length=8K2026.06 | 48.6 | |
| FlowTracerBackbone=Qwen3-8B-Base, Context Length=1K2026.06 | 46.7 | |
| High-entropyBackbone=Qwen3-8B-Base, Context Length=1K2026.06 | 45.9 | |
| AttentionBackbone=Qwen3-8B-Base, Context Length=1K2026.06 | 45.9 | |
| GradientBackbone=Qwen3-8B-Base, Context Length=1K2026.06 | 45 | |
| CorrelationBackbone=Qwen3-8B-Base, Context Length=1K2026.06 | 44.8 | |
| FlowTracerBackbone=Qwen3-4B-Base, Context Length=1K2026.06 | 44.2 | |
| GRPOBackbone=Qwen3-8B-Base, Context Length=1K2026.06 | 44.2 | |
| GradientBackbone=Qwen3-4B-Base, Context Length=1K2026.06 | 43.3 | |
| RandomBackbone=Qwen3-8B-Base, Context Length=1K2026.06 | 43.3 | |
| AttentionBackbone=Qwen3-4B-Base, Context Length=1K2026.06 | 43.1 | |
| GRPOBackbone=Qwen3-4B-Base, Context Length=1K2026.06 | 42.8 | |
| High-entropyBackbone=Qwen3-4B-Base, Context Length=1K2026.06 | 42.8 | |
| RandomBackbone=Qwen3-4B-Base, Context Length=1K2026.06 | 42 | |
| CorrelationBackbone=Qwen3-4B-Base, Context Length=1K2026.06 | 42 | |
| PAPOBase Model=Qwen2.5-14B-Base2026.05 | 39.22 | |
| DAPOBase Model=Qwen2.5-14B-Base2026.05 | 38.57 | |
| Ent-RegBase Model=Qwen2.5-14B-Base2026.05 | 38.34 | |
| 80/20Base Model=Qwen2.5-14B-Base2026.05 | 37.59 | |
| PAPOBase Model=Qwen2.5-7B-Base2026.05 | 34.7 | |
| DAPOBase Model=Qwen2.5-7B-Base2026.05 | 33.12 | |
| Ent-RegBase Model=Qwen2.5-7B-Base2026.05 | 33.08 | |
| 80/20Base Model=Qwen2.5-7B-Base2026.05 | 32.69 | |
| BaseBase Model=Qwen2.5-14B-Base2026.05 | 26.04 | |
| BaseBase Model=Qwen2.5-7B-Base2026.05 | 22.3 |