DeepSeek V4.1 Flash 1017 scores 38.7, beating Opus 5.5's 31.7 in bug-hunt benchmark

siegl_rene · x · 2026-10-19

After removing a cheating sample from his Bug Hunt benchmark, Pawel Huryn reports updated scores: DeepSeek V4.1 Flash 1017 (max, n=3) at 38.7 and (high, n=2) at 32, versus Opus 5.5 (high, n=3) at 31.7 — calling the value/cost curve "crazy". Respondent sieglrene cautions that with only 2-3 runs a single unusually good run can skew rankings, and asks whether 1017 stays competitive at the median.

Original post →