Mathematical Reasoning on GSM8K (Acc., Final Gap)
81.2AccuracyPioneer Agent
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Pioneer AgentModel=Qwen3-8B, System Strategy=Agent, Checkpoint Stage=Best2026.04 | 81.2 | 16.5 | |
| Pioneer AgentModel=Qwen3-8B, System Strategy=Agent, Checkpoint Stage=C12026.04 | 75.9 | 16.5 | |
| Naive BaselineModel=Qwen3-8B, System Strategy=Naive, Checkpoint Stage=C12026.04 | 71.6 | 16.5 | |
| Naive BaselineModel=Qwen3-8B, System Strategy=Naive, Checkpoint Stage=C42026.04 | 64.7 | 16.5 | |
| Pioneer AgentModel=Llama 3.2-3B, System Strategy=Agent, Checkpoint Stage=Best2026.04 | 34.8 | 20.3 | |
| Pioneer AgentModel=Llama 3.2-3B, System Strategy=Agent, Checkpoint Stage=C12026.04 | 27.8 | 20.3 | |
| Naive BaselineModel=Llama 3.2-3B, System Strategy=Naive, Checkpoint Stage=C12026.04 | 19.8 | 20.3 | |
| Naive BaselineModel=Llama 3.2-3B, System Strategy=Naive, Checkpoint Stage=C42026.04 | 14.5 | 20.3 |