Google · Proprietary · tested 10 Oct 2026

Gemini 3.8 Flash High effort

Overall rank#16 of 25

Safety refusals affected this result

Google's cyber safeguards stopped attempts in 71 of 100 labs. Stopped attempts score zero.

Its cost, tokens and time don't cover the full benchmark, so they're left out of comparisons. How refusals are scored

Solves37/10021 first attempt · +16 retry

Rungs proven178/400109 first attempt · +69 retry

Cost to runn/aNot comparable: safety refusals

Tokens usedn/aNot comparable: safety refusals

Time per labn/aNot comparable: safety refusals

Where it places

Each line is one metric across all 25 configurations. Right is always better; this model is the large dot.

Solveslabs solved, of 100

worsebetter

37#16 of 25

Rungs provenrungs proven, of 400

worsebetter

178#20 of 25

First-attempt solvessolved on the first attempt

worsebetter

21#20 of 25

Cost per solveUSD per solved lab

worsebetter

n/anot ranked

How Gemini 3.8 Flash (High) did on each type of vulnerability. Shown as counts, because some classes have only a few labs.

Authentication / authorisation bypassSafety refusals in 22 of 28 labs10 of 28 solved50 of 112 rungs
Information disclosureSafety refusals in 7 of 12 labs5 of 12 solved21 of 48 rungs
IDORSafety refusals in 10 of 11 labs2 of 11 solved14 of 44 rungs
SSRFSafety refusals in 8 of 10 labs1 of 10 solved9 of 40 rungs
Business logicSafety refusals in 6 of 10 labs3 of 10 solved14 of 40 rungs
Key or secret exposureSafety refusals in 3 of 8 labs7 of 8 solved28 of 32 rungs
XSSSafety refusals in 4 of 6 labs3 of 6 solved13 of 24 rungs
Token theftSafety refusals in 2 of 5 labs5 of 5 solved20 of 20 rungs

Fewer than 5 labs each: one lab changes these a lot, so read them with care.

Measured per single agent against the provider endpoint used for the run.

Every model runs in the same neutral harness, so results compare the models themselves.