Multi-step mathematical reasoning on GSM8K (test)
96.61AccuracyTCP-MCP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TCP-MCPBackbone=DeepSeek-V3.22026.05 | 96.61 | 6.02 | |
| PHPBackbone=DeepSeek-V3.22026.05 | 96.21 | 0.71 | |
| ChatDevBackbone=DeepSeek-V3.22026.05 | 96.13 | 0.44 | |
| Self-ConsistencyBackbone=DeepSeek-V3.22026.05 | 96.05 | 1.96 | |
| G-DesignerBackbone=DeepSeek-V3.22026.05 | 95.81 | 4.44 | |
| Random GraphBackbone=DeepSeek-V3.22026.05 | 95.81 | 14.78 | |
| Zero-shotBackbone=DeepSeek-V3.22026.05 | 95.78 | 0.21 | |
| LLM-DebateBackbone=DeepSeek-V3.22026.05 | 95.48 | 24.61 | |
| Complex CoTBackbone=DeepSeek-V3.22026.05 | 94.18 | 0.45 | |
| MetaGPTBackbone=DeepSeek-V3.22026.05 | 93.06 | 0.12 | |
| LLM-BlenderBackbone=DeepSeek-V3.22026.05 | 92.87 | 0.33 | |
| Complete GraphBackbone=DeepSeek-V3.22026.05 | 92.11 | 7.65 | |
| AutoGenBackbone=DeepSeek-V3.22026.05 | 88.79 | 0.39 | |
| GPTSwarmBackbone=DeepSeek-V3.22026.05 | 77.71 | 1.21 |