List your company

Mathematical Problem Solving on MATH (Accuracy & Cost)

95.7AccuracyOracle (Perfect Knowledge)
-1.12424.01349.1574.287Feb 10, 2026Feb 21, 2026Mar 4, 2026Mar 15, 2026Mar 26, 2026Apr 6, 2026Apr 18, 2026

Evaluation results

MethodLinks
Oracle (Perfect Knowledge)Model=ensemble2026.0295.76.37
Probe Routerlambda=0, Model=ensemble2026.029328.37
SFTModel=DS-R1-Qwen-32B2026.0492.65—
D-QRELOModel=DS-R1-Qwen-32B2026.0492.45—
SFTModel=DS-R1-Qwen-14B2026.0492.25—
gpt-oss-20b-highModel=gpt-oss-20b (high)2026.029240
Probe Routerlambda=0.2, Model=ensemble2026.0291.710.35
gpt-oss-20b-mediumModel=gpt-oss-20b (medium)2026.0291.411.78
R1-Qwen-7BModel=DeepSeek-R1-Distill-Qwen-7B2026.0291.418.43
D-QRELOModel=DS-R1-Qwen-14B2026.0491.4—
Probe Routerlambda=0.4, Model=ensemble2026.0290.98.29
Probe Routerlambda=0.6, Model=ensemble2026.0290.27.02
SFTModel=DS-R1-LLaMA-8B2026.0489.55—
Probe Routerlambda=0.8, Model=ensemble2026.0289.45.98
Probe Routerlambda=1, Model=ensemble2026.02895.34
Random RoutingModel=ensemble2026.0288.915.43
Delta-CoMeModel=DS-R1-Qwen-32B2026.0488.1—
Delta-CoMeModel=DS-R1-Qwen-14B2026.0487.7—
gpt-oss-20b-lowModel=gpt-oss-20b (low)2026.0286.64
Math-7BModel=Qwen2.5-Math-7B-Instruct2026.0282.73.28
D-QRELOModel=DS-R1-LLaMA-8B2026.0479.05—
MagnitudeModel=DS-R1-Qwen-32B2026.0478.65—
WandaModel=DS-R1-Qwen-32B2026.0478.4—
WandaModel=DS-R1-Qwen-14B2026.0478.2—
MagnitudeModel=DS-R1-Qwen-14B2026.0476.35—
BackboneModel=DS-R1-Qwen-14B2026.0464.15—
BackboneModel=DS-R1-Qwen-32B2026.0462.4—
SVDModel=DS-R1-Qwen-32B2026.0460.3—
RandomModel=DS-R1-Qwen-14B2026.0458.45—
RandomModel=DS-R1-Qwen-32B2026.0457.3—
SVDModel=DS-R1-Qwen-14B2026.0453.9—
Delta-CoMeModel=DS-R1-LLaMA-8B2026.0445.4—
SVDModel=DS-R1-LLaMA-8B2026.0439.2—
WandaModel=DS-R1-LLaMA-8B2026.0424.7—
BitDeltaModel=DS-R1-Qwen-32B2026.0422.75—
BitDeltaModel=DS-R1-Qwen-14B2026.0420.95—
MagnitudeModel=DS-R1-LLaMA-8B2026.0418.4—
BitDeltaModel=DS-R1-LLaMA-8B2026.0411.8—
RandomModel=DS-R1-LLaMA-8B2026.047.65—
BackboneModel=DS-R1-LLaMA-8B2026.042.6—