List your company

Code Generation on HumanEval (Accuracy, Mean, Drop)

94.05AccuracyBF16
-3.76221.631547.02572.4185May 18, 2026

Evaluation results

MethodLinks
BF16Model=Qwen3-4B Thinking-2507, BPE=16.002026.0594.0575.64—
OSCARModel=Qwen3-4B Thinking-2507, BPE=2.282026.0592.2471.864-3.78
BF16Model=GLM-4.7 FP8, 358B, BPE=16.002026.0591.4677.89—
Saw-INT4Model=GLM-4.7 FP8, 358B, BPE=4.252026.0591.377.950.06
BF16Model=Qwen3-32B, BPE=16.002026.0591.1974.19—
OSCARModel=GLM-4.7 FP8, 358B, BPE=2.282026.0591.0678.160.27
Saw-INT4Model=Qwen3-32B, BPE=4.252026.0590.8574.430.24
TurboQuant (no MP)Model=GLM-4.7 FP8, 358B, BPE=3.252026.0590.2478.150.26
OSCARModel=Qwen3-32B, BPE=2.282026.0590.1274.17-0.02
QuaRot-INT2Model=GLM-4.7 FP8, 358B, BPE=2.252026.0589.8775.14-2.75
Saw-INT4Model=Qwen3-4B Thinking-2507, BPE=4.252026.0589.7873.11-2.53
TurboQuant (no MP)Model=Qwen3-32B, BPE=3.252026.0588.4171.99-2.2
OSCARModel=Qwen3-8B, BPE=2.282026.0587.8869.416-1.42
Naive INT2Model=GLM-4.7 FP8, 358B, BPE=2.252026.0586.560.49-17.4
Saw-INT4Model=Qwen3-8B, BPE=4.252026.0586.4469.97-0.87
BF16Model=Qwen3-8B, BPE=16.002026.0585.9570.84—
TurboQuant (no MP)Model=Qwen3-8B, BPE=3.252026.0574.6356.88-13.96
TurboQuant (no MP)Model=Qwen3-4B Thinking-2507, BPE=3.252026.0531.8331.74-43.9
QuaRot-INT2Model=Qwen3-8B, BPE=2.252026.059.810.14-60.7
QuaRot-INT2Model=Qwen3-32B, BPE=2.252026.051.837.9-66.29
QuaRot-INT2Model=Qwen3-4B Thinking-2507, BPE=2.252026.050.981.4-74.24
Naive INT2Model=Qwen3-4B Thinking-2507, BPE=2.252026.0500-75.64
Naive INT2Model=Qwen3-8B, BPE=2.252026.0500-70.84
Naive INT2Model=Qwen3-32B, BPE=2.252026.0500-74.19