List your company

Multi-step mathematical reasoning on GSM8K (test)

96.61AccuracyTCP-MCP
76.95482.05787.1692.263May 27, 2026

Evaluation results

MethodLinks
TCP-MCPBackbone=DeepSeek-V3.22026.0596.616.02
PHPBackbone=DeepSeek-V3.22026.0596.210.71
ChatDevBackbone=DeepSeek-V3.22026.0596.130.44
Self-ConsistencyBackbone=DeepSeek-V3.22026.0596.051.96
G-DesignerBackbone=DeepSeek-V3.22026.0595.814.44
Random GraphBackbone=DeepSeek-V3.22026.0595.8114.78
Zero-shotBackbone=DeepSeek-V3.22026.0595.780.21
LLM-DebateBackbone=DeepSeek-V3.22026.0595.4824.61
Complex CoTBackbone=DeepSeek-V3.22026.0594.180.45
MetaGPTBackbone=DeepSeek-V3.22026.0593.060.12
LLM-BlenderBackbone=DeepSeek-V3.22026.0592.870.33
Complete GraphBackbone=DeepSeek-V3.22026.0592.117.65
AutoGenBackbone=DeepSeek-V3.22026.0588.790.39
GPTSwarmBackbone=DeepSeek-V3.22026.0577.711.21