List your company
ResearchFormerly WizWand

Scientific Problem Solving on SciBench (Accuracy, Δ)

72.1Accuracyo3
44.95652.00359.0566.097Aug 26, 2025
Updated 5mo ago

Evaluation results

MethodLinks
o3Reasoning Effort=High2025.0872.12.4
GPT-5Reasoning Effort=High2025.08721.6
Gemini-2.5-ProReasoning Effort=Low2025.0871—
GPT-5Reasoning Effort=Low2025.0870.4—
Gemini-2.5-ProReasoning Effort=High2025.0870.2-0.8
o3Reasoning Effort=Low2025.0869.7—
o4-miniReasoning Effort=High2025.0869.74.2
Claude-Sonnet-4Reasoning Effort=High2025.0867.11.6
o3-miniReasoning Effort=High2025.0866.320.3
o4-miniReasoning Effort=Low2025.0865.5—
Claude-Sonnet-4Reasoning Effort=Low2025.0865.5—
o3-miniReasoning Effort=Low2025.0846—