List your company
ResearchFormerly WizWand

General Model Evaluation on Comprehensive Reasoning Suite (MMLU, HumanEval, GSM8K, GPQA, etc.)

69.3Average ScoreCHORD
37.247245.568653.8962.2114Sep 29, 2025
Updated 4mo ago

Evaluation results

MethodLinks
CHORDBackbone=Qwen3-4B2025.0969.3
A-CHORDBackbone=Qwen3-4B2025.0969.27
A-GRPOBackbone=Qwen3-4B2025.0968.39
GRPOBackbone=Qwen3-4B2025.0968.11
SFT → A-GRPOBackbone=Qwen3-4B2025.0968.11
A-SFT → A-GRPOBackbone=Qwen3-4B2025.0967.93
SFT → GRPOBackbone=Qwen3-4B2025.0965.75
Qwen3-4BBackbone=Qwen3-4B2025.0962.52
SFTBackbone=Qwen3-4B2025.0960.8
A-SFTBackbone=Qwen3-4B2025.0960.67
SFT → A-GRPOBackbone=Qwen3-0.6B2025.0943.61
A-SFT → A-GRPOBackbone=Qwen3-0.6B2025.0943.31
SFT → GRPOBackbone=Qwen3-0.6B2025.0942.56
A-CHORDBackbone=Qwen3-0.6B2025.0942.09
A-GRPOBackbone=Qwen3-0.6B2025.0941.99
CHORDBackbone=Qwen3-0.6B2025.0941.19
GRPOBackbone=Qwen3-0.6B2025.0940.93
Qwen3-0.6BBackbone=Qwen3-0.6B2025.0940.81
A-SFTBackbone=Qwen3-0.6B2025.0939.83
SFTBackbone=Qwen3-0.6B2025.0938.48