benchgap
StepFun · model

Step 3.5 Flash benchmark scores

As of 2026-10-10, Step 3.5 Flash (StepFun) has measured scores on 3 benchmarks and estimated scores on 16 more.

BenchmarkScoreSource
AA-SciCode45.4%estimated ± 3.9 pp, high confidence
AIME2696.5%estimated ± 4.6 pp, low confidence
LiveCodeBench v686.4%measured
SciCode45.9%estimated ± 3.8 pp, high confidence
AA Coding Index49.1%estimated ± 5.5 pp, medium confidence
SWE-bench Verified74.4%measured
SWE-bench Pro50.1%estimated ± 3.4 pp, high confidence
FrontierMath v2 (Tier 4)4.9%estimated ± 6.6 pp, low confidence
AIME 202597.3%measured
FrontierCode 1.1 Main9.5%estimated ± 2.7 pp, low confidence
Vals LiveCodeBench84.1%estimated ± 2.0 pp, low confidence
Vals SWE-bench68.7%estimated ± 6.8 pp, medium confidence
Vibe Code Bench2.4%estimated ± 8.1 pp, medium confidence
HMMT Feb 202598.1%estimated ± 3.1 pp, medium confidence
NL2Repo38.4%estimated ± 3.4 pp, high confidence
React Native Evals73.2%estimated ± 2.3 pp, medium confidence
SWE-Rebench57.1%estimated ± 3.6 pp, medium confidence
AIME 202496.3%estimated ± 2.2 pp, low confidence
MATH-50098.2%estimated ± 2.9 pp, medium confidence