Deep search QA on Frames
46.42AccuracyProCeedRL
Evaluation Results
| Method | Links | |
|---|---|---|
| ProCeedRLBackbone=Qwen3-8B2026.04 | 46.42 | |
| DAPO/Search-R1Backbone=Qwen3-8B2026.04 | 43.59 | |
| RFTBackbone=Qwen3-8B2026.04 | 40.65 | |
| Rewinding MoreBackbone=Qwen3-8B, Type=ablation2026.04 | 40.13 | |
| ECHOBackbone=Qwen3-32B-Instruct2026.06 | 39.1 | |
| Qwen3-8B-v3-SFTBackbone=Qwen3-8B2026.04 | 37.5 | |
| ECHO w/ Top-K retrievalBackbone=Qwen3-32B-Instruct2026.06 | 37.3 | |
| ECHO w/o Traceable CA & Turn-level ISBackbone=Qwen3-32B-Instruct2026.06 | 32.2 | |
| ReAct PromptingBackbone=Qwen3-8B2026.04 | 31.92 | |
| ECHO w/ Top-K retrieval & w/o turn summaryBackbone=Qwen3-32B-Instruct2026.06 | 31.3 | |
| ECHO w/o traceable CABackbone=Qwen3-32B-Instruct2026.06 | 30.8 | |
| SUPOBackbone=Qwen3-32B-Instruct2026.06 | 26.8 | |
| ECHOBackbone=Qwen3-30B-A3B-Instruct2026.06 | 25 | |
| GRPOBackbone=Qwen3-32B-Instruct2026.06 | 24.8 | |
| GRPOBackbone=Qwen3-30B-A3B-Instruct2026.06 | 19.1 | |
| SUPOBackbone=Qwen3-30B-A3B-Instruct2026.06 | 17 |