List your company

Mathematical Reasoning on MATH500 (Accuracy, Token Usage, and Reduction)

61AccuracyReasoning
53.30455.30257.359.298Jan 28, 2026

Evaluation results

MethodLinks
ReasoningModel=Qwen3-8B2026.01614,150—
KV-ClassificationModel=Qwen3-8B2026.0160.43,963-4.5
ReasoningModel=DeepSeek-R1-14B2026.01591,839—
KV-ClassificationModel=DeepSeek-R1-14B2026.0157.81,506-18.1
KV-GenerativeModel=Qwen3-8B2026.0157.8727-82.5
Fast ThinkingModel=Qwen3-8B2026.0156.8616—
KV-GenerativeModel=DeepSeek-R1-14B2026.0156.6657-64.3
Fast ThinkingModel=DeepSeek-R1-14B2026.0153.6540—