Real-data Math
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Real-data Math Mistral outputs (3% verified data)
50.46Original Accuracy
6
Real-data Math Qwen outputs (3% verified data)
55.1Original Accuracy
6
Real-data Math Gemini outputs (3% verified data)
66.35Original Accuracy
6
Real-data Math GPT-5.4-mini (3% verified data)
62.35Original Accuracy
6
Real-data Math GPT-5.4 outputs (3% verified data)
67.2Original Accuracy
6