Model

Claude 3 Sonnet

Anthropic · closed weights

anthropic.com

Knowledge & Reasoning

Knowledge & Reasoning
BenchmarkScoreMetricHarnessSourceConfidenceLink
GPQA Diamond40.6accuracyN/AIndependenthighepoch.ai
MMLU79.0average_(%)N/AIndependentmediumanthropic.com
MMLU75.9exact_matchN/AIndependentmediumcrfm.stanford.edu
WinoGrande75.1accuracyN/ASelf-reportedmediumassets.anthropic.com
WinoGrande75.1accuracyN/AIndependentmediumanthropic.com

Coding

Coding
BenchmarkScoreMetricHarnessSourceConfidenceLink
MBPP79.4accuracyN/AIndependentmediumanthropic.com

Multimodal

Multimodal
BenchmarkScoreMetricHarnessSourceConfidenceLink
VideoDB's OCR Benchmark Public Collection67.7average_accuracyN/AIndependentmediumarxiv.org

Safety & Alignment

Safety & Alignment
BenchmarkScoreMetricHarnessSourceConfidenceLink
SynthPAI70.9average_accuracy_in_%N/AIndependentmediumarxiv.org

Reproduced by the community

Loading community runs…