Tool-using Reasoning on Reasoning Domain Suite (AIME2024, AIME2025, HotpotQA, 2WikiMultihopQA, Musique)
42.39Average AccuracyTEPOdense
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TEPOdenseBackbone=Qwen2.5-7B-Instruct2026.02 | 42.39 | 3.99 | |
| TEPOdenseBackbone=Llama3.1-8B-Instruct2026.02 | 39.01 | 3.6 | |
| GRPOBackbone=Qwen2.5-7B-Instruct2026.02 | 38.93 | 1.97 | |
| ARPOBackbone=Qwen2.5-7B-Instruct2026.02 | 37.92 | 1.67 | |
| AEPOBackbone=Qwen2.5-7B-Instruct2026.02 | 37.9 | 2.96 | |
| TEPOsparseBackbone=Qwen2.5-7B-Instruct2026.02 | 37.77 | 1.64 | |
| ARPOBackbone=Llama3.1-8B-Instruct2026.02 | 37.37 | 1.79 | |
| SFTBackbone=Qwen2.5-7B-Instruct2026.02 | 37.2 | 3.98 | |
| TEPOsparseBackbone=Llama3.1-8B-Instruct2026.02 | 36.43 | 1.66 | |
| GRPOBackbone=Llama3.1-8B-Instruct2026.02 | 36.39 | 2.02 | |
| SFTBackbone=Llama3.1-8B-Instruct2026.02 | 31.74 | 5.42 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct2026.02 | 25.48 | 1.98 | |
| Llama3.1-8B-InstructBackbone=Llama3.1-8B-Instruct2026.02 | 19.25 | 3.45 |