Platform benchmark

deepseek-v4-flash-0731

deepseek · MATH-500 · seed 42 · 2026-10-01 06:06 UTC

ResultsMATH-500 · n=96 · seed 42
93.8%

93.8% correct · 100.0% answered on MATH-500

Score per $
18.13
Run cost
$0.05
Latency p50
3.1 s
Latency p95
14.3 s
Items
96/96 completed
Dropped
0

Dataset: MATH-500 · per-item question and response text available in the results.