Reasoning on BBEH
15.7pass@1DenoiseRL-DAPO
Evaluation Results
| Method | Links | |
|---|---|---|
| DenoiseRL-DAPOBase Model=Qwen3-4B-Base2026.05 | 15.7 | |
| Qwen3-8BBackbone=Qwen3-8B2026.01 | 15.31 | |
| DenoiseRL-GRPOBase Model=Qwen3-4B-Base2026.05 | 14.8 | |
| RelayLLM (Simple)Student Model Backbone=Qwen3-1.7B2026.01 | 12.67 | |
| DenoiseRL-DAPOBase Model=Qwen3-8B-Base2026.05 | 12.6 | |
| RelayLLM (Difficulty-Aware)Student Model Backbone=Qwen3-1.7B2026.01 | 12.46 | |
| DAPOBase Model=Qwen3-8B-Base2026.05 | 11.7 | |
| CITERStudent Model Backbone=Qwen3-1.7B2026.01 | 11.67 | |
| DenoiseRL-GRPOBase Model=Qwen3-8B-Base2026.05 | 11.5 | |
| GRPOBase Model=Qwen3-4B-Base2026.05 | 11.1 | |
| GRPOStudent Model Backbone=Qwen3-1.7B2026.01 | 10.89 | |
| GRPOBase Model=Qwen3-8B-Base2026.05 | 10.6 | |
| DAPOBase Model=Qwen3-4B-Base2026.05 | 10.4 | |
| Base ModelStudent Model Backbone=Qwen3-1.7B2026.01 | 9.91 | |
| RelayLLM (Difficulty-Aware)Student Model Backbone=Qwen3-0.6B2026.01 | 8.56 | |
| RelayLLM (Simple)Student Model Backbone=Qwen3-0.6B2026.01 | 8.32 | |
| CITERStudent Model Backbone=Qwen3-0.6B2026.01 | 8.16 | |
| GRPOStudent Model Backbone=Qwen3-0.6B2026.01 | 7.82 | |
| Base ModelStudent Model Backbone=Qwen3-0.6B2026.01 | 7.19 | |
| GTPOBase Model=Qwen2.5-7B2025.11 | 4.4 | |
| GRPOBase Model=Qwen2.5-7B2025.11 | 4.1 | |
| BaseBase Model=Qwen3-4B-Base2026.05 | 4.1 | |
| BaseBase Model=Qwen3-8B-Base2026.05 | 4.1 |