Multi-hop Question Answering on HotpotQA (Score, Relative Gain)
80.3Score+PACE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| +PACESLM=Qwen3.5-9B2026.05 | 80.3 | 6.5 | |
| +PESLM=Qwen3.5-9B2026.05 | 78.9 | 4.6 | |
| Gödel AgentSLM=Qwen3.5-9B2026.05 | 78 | 3.5 | |
| DeepSeek V3.2SLM=DeepSeek V3.22026.05 | 77.9 | — | |
| +PACESLM=Qwen3-4B2026.05 | 77.6 | 0 | |
| VanillaSLM=Qwen3-4B2026.05 | 77.5 | — | |
| GEPASLM=Qwen3-4B2026.05 | 77.5 | 0 | |
| ACESLM=Qwen3-4B2026.05 | 77.5 | 0 | |
| +PESLM=Qwen3-4B2026.05 | 77.5 | 0 | |
| +PACESLM=Ministral-3-14B2026.05 | 77.4 | 29.8 | |
| Gödel AgentSLM=Qwen3-4B2026.05 | 77.2 | -0.5 | |
| ACESLM=Qwen3.5-9B2026.05 | 77.1 | 2.3 | |
| +CESLM=Qwen3-4B2026.05 | 77 | -0.1 | |
| MIPROv2SLM=Qwen3-4B2026.05 | 75.8 | -2.1 | |
| VanillaSLM=Qwen3.5-9B2026.05 | 75.4 | — | |
| +CESLM=Qwen3.5-9B2026.05 | 75.2 | -0.2 | |
| GEPASLM=Qwen3.5-9B2026.05 | 74.1 | -1.7 | |
| MIPROv2SLM=Qwen3.5-9B2026.05 | 73.9 | -2 | |
| +PESLM=Ministral-3-14B2026.05 | 70.9 | 18.9 | |
| Gödel AgentSLM=Ministral-3-14B2026.05 | 65.9 | 10.6 | |
| +CESLM=Ministral-3-14B2026.05 | 65.4 | 9.7 | |
| GEPASLM=Ministral-3-14B2026.05 | 62.5 | 4.8 | |
| ACESLM=Ministral-3-14B2026.05 | 60.5 | 1.5 | |
| VanillaSLM=Ministral-3-14B2026.05 | 59.6 | — | |
| MIPROv2SLM=Ministral-3-14B2026.05 | 58.3 | -2.2 |