Multi-turn agent task on ACEBench multi-turn (test)
76.5Process AccuracyH-EPM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| H-EPMModel=GPT-4.12025.12 | 76.5 | 56.7 | |
| H-EPMModel=GPT-4O2025.12 | 72 | 50.3 | |
| H-EPMModel=GPT-4.1-MINI2025.12 | 71.9 | 53.3 | |
| Reasoning BankModel=GPT-4.12025.12 | 70.3 | 49.8 | |
| FAMA-BaseBackbone=Qwen2.5-72B-Instruct2026.04 | 70.2 | 50 | |
| Reasoning BankModel=GPT-4.1-MINI2025.12 | 70.1 | 48.3 | |
| Reasoning BankModel=GPT-4O2025.12 | 68.4 | 47.7 | |
| BaseModel=GPT-4O2025.12 | 66.4 | 48.4 | |
| ToolNetModel=GPT-4.12025.12 | 66.4 | 48.1 | |
| ToolNetModel=GPT-4O2025.12 | 65.9 | 47.3 | |
| BaseModel=GPT-4.12025.12 | 65.3 | 50.3 | |
| FAMA-ReActBackbone=Qwen2.5-72B-Instruct2026.04 | 65.2 | 43.3 | |
| Retrieval Top-1Model=GPT-4.1-MINI2025.12 | 64.1 | 47.3 | |
| BaseModel=GPT-4.1-MINI2025.12 | 63.3 | 40 | |
| Retrieval Top-1Model=GPT-4O2025.12 | 62.8 | 46.7 | |
| ToolNetModel=GPT-4.1-MINI2025.12 | 62.7 | 43.6 | |
| ReActBackbone=Qwen3-14B2026.04 | 61.3 | 20 | |
| Retrieval Top-1Model=GPT-4.12025.12 | 60.1 | 43.2 | |
| FAMA-ReActBackbone=Qwen3-32B2026.04 | 59.6 | 30 | |
| FAMA-ReActBackbone=Qwen3-14B2026.04 | 58.4 | 23.3 | |
| ReActBackbone=Qwen2.5-72B-Instruct2026.04 | 57.6 | 36.7 | |
| FAMA-ReActBackbone=Qwen3-4B-Instruct2026.04 | 53.9 | 20 | |
| ReActBackbone=Qwen3-4B-Instruct2026.04 | 51.3 | 10 | |
| ReActBackbone=Qwen3-32B2026.04 | 50.2 | 13.3 | |
| FAMA-BaseBackbone=Qwen3-14B2026.04 | 50 | 23.3 | |
| FAMA-BaseBackbone=Qwen3-32B2026.04 | 50 | 23.3 | |
| BaseBackbone=Qwen3-32B2026.04 | 46.5 | 23.3 | |
| FAMA-BaseBackbone=Qwen3-4B-Instruct2026.04 | 45.1 | 16.7 | |
| BaseBackbone=Qwen2.5-72B-Instruct2026.04 | 45.1 | 23.3 | |
| BaseBackbone=Qwen3-4B-Instruct2026.04 | 41.6 | 10 | |
| BaseBackbone=Qwen3-14B2026.04 | 40.8 | 16.7 |