Multi-Answer Question Answering on Mintaka
64.2PrecisionSPADER
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SPADERBackbone=Llama3.1-8B2026.05 | 64.2 | 64.4 | 62.5 | |
| StepSearchBackbone=Llama3.1-8B2026.05 | 63.9 | 61.2 | 61.8 | |
| SPADERBackbone=Qwen3-8B2026.05 | 62.5 | 62.6 | 61.1 | |
| R3-RAGBackbone=Llama3.1-8B2026.05 | 61.7 | 63 | 60.7 | |
| PPOBackbone=Llama3.1-8B2026.05 | 61 | 59.5 | 60 | |
| R3-RAGBackbone=Qwen3-8B2026.05 | 58.7 | 55.9 | 56.3 | |
| StepSearchBackbone=Qwen3-8B2026.05 | 58.5 | 56.2 | 56.5 | |
| PPOBackbone=Qwen3-8B2026.05 | 57.3 | 54.6 | 55 | |
| GRPOBackbone=Llama3.1-8B2026.05 | 54 | 52.7 | 52.7 | |
| SPADER (QAMPARI-only)Backbone=Llama3.1-8B2026.05 | 53.8 | 59.4 | 55.8 | |
| GRPOBackbone=Qwen3-8B2026.05 | 52.8 | 49.7 | 50.4 | |
| SPADER (QAMPARI-only)Backbone=Qwen3-8B2026.05 | 51.1 | 53.6 | 51.3 | |
| ReActBackbone=Llama3.1-8B2026.05 | 40.9 | 48.1 | 42.4 | |
| ReActBackbone=Qwen3-8B2026.05 | 38.4 | 46.6 | 40.4 | |
| RAGBackbone=Llama3.1-8B2026.05 | 33 | 42.5 | 35.5 | |
| RAGBackbone=Qwen3-8B2026.05 | 29.7 | 44.8 | 32.5 |