benchgap
StepFun · model

Step 5 Preview benchmark scores

As of 2026-10-07, Step 5 Preview (StepFun) has measured scores on 13 benchmarks and estimated scores on 3 more.

BenchmarkScoreSource
Terminal-Bench 2.187.4%estimated ± 4.0 pp, high confidence
Terminal-Bench 4.033.3%measured
Terminal-Bench Science 0.12.4%measured
GPQA Diamond93.3%estimated ± 1.3 pp, high confidence
SciCode58.9%measured
CritPt20.9%measured
GDP.pdf14.8%measured
τ³-Bench Banking44.4%estimated ± 2.6 pp, low confidence
MMMU-Pro76.0%measured
AA-AnalystAgent35.0%measured
Harvey LAB93.4%measured
Humanity's Last Exam46.5%measured
AutomationBench51.0%measured
EnterpriseOps-Gym47.2%measured
APEX-Agents38.0%measured
ITBench SRE55.6%measured