List your company

Mathematical Reasoning on MATH-500 (Score, Retention)

96.4ScoreHaiku 4.5
57.19267.37177.5587.729Mar 30, 2026

Evaluation results

MethodLinks
Haiku 4.5Training Phase=Original (Orig)2026.0396.4—
Haiku 4.5Training Phase=Trojan-Speak (TS)2026.0393.397
Qwen3 14BTraining Phase=Original (Orig)2026.0387.1—
Qwen3 32BTraining Phase=Original (Orig)2026.0386.9—
Qwen3 4BTraining Phase=Original (Orig)2026.0384.3—
Qwen3 8BTraining Phase=Original (Orig)2026.0382.8—
Qwen3 32BTraining Phase=Trojan-Speak (TS)2026.0382.495
Qwen3 14BTraining Phase=Trojan-Speak (TS)2026.0379.492
Qwen3 8BTraining Phase=Trojan-Speak (TS)2026.0364.378
Qwen3 4BTraining Phase=Trojan-Speak (TS)2026.0358.770