ResearchBenchmarksMassive Multitask Language Understanding on MMLU Pro (Log Accuracy and Efficiency)Follow38.6Normalized Log AccuracyT-Free36.83237.29137.7538.209Mar 16, 2026Evaluation ResultsMethodMethodLinksNormalized Log AccuracyMemory Usage (Bytes/Position)T-Freeshots=5-shotshots=5-shot2026.0338.64.73HATifiedshots=5-shotshots=5-shot2026.0336.94.73Llamashots=5-shotshots=5-shot2026.0336.93.73