List your company
ResearchFormerly WizWand

Scientific Question Answering on GPQA (Accuracy (%), Δ)

82.4AccuracyGPT-5
62.6467.7772.978.03Aug 26, 2025
Updated 5mo ago

Evaluation results

MethodLinks
GPT-5Reasoning Effort=High2025.0882.43.1
Gemini-2.5-ProReasoning Effort=Low2025.0880.1—
o3Reasoning Effort=High2025.0879.94.5
Gemini-2.5-ProReasoning Effort=High2025.0879.5-0.6
GPT-5Reasoning Effort=Low2025.0879.2—
o3Reasoning Effort=Low2025.0875.4—
o4-miniReasoning Effort=High2025.0874.65.2
o3-miniReasoning Effort=High2025.0873.910.5
o4-miniReasoning Effort=Low2025.0869.4—
Claude-Sonnet-4Reasoning Effort=High2025.08695.2
Claude-Sonnet-4Reasoning Effort=Low2025.0863.8—
o3-miniReasoning Effort=Low2025.0863.4—