OpenAI · Proprietary · tested 27 Oct 2026

GPT-6.1 Terra Max effort

Overall rank#24 of 30

Safety refusals affected this result

OpenAI's cyber safeguards stopped attempts in 89 of 100 labs. Stopped attempts score zero. BB Arena is enrolled in OpenAI's cyber program for verified security researchers, and these refusals happened anyway.

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

Solves10/1005 first attempt · +5 retry

Rungs proven41/40022 first attempt · +19 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 30 configurations. Right is always better; this model is the large dot.

Solveslabs solved, of 100

worsebetter

10#24 of 30

Rungs provenrungs proven, of 400

worsebetter

41#25 of 30

First-attempt solvessolved on the first attempt

worsebetter

5#24 of 30

Cost per solveUSD per solved lab

worsebetter

n/anot ranked

By vulnerability class

How GPT-6.1 Terra (Max) 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 labs7 of 28 solved29 of 112 rungs
Information disclosureSafety refusals in 12 of 12 labs1 of 12 solved4 of 48 rungs
IDORSafety refusals in 11 of 11 labs1 of 11 solved4 of 44 rungs
SSRFSafety refusals in 10 of 10 labs0 of 10 solved0 of 40 rungs
Business logicSafety refusals in 10 of 10 labs0 of 10 solved0 of 40 rungs
Key or secret exposureSafety refusals in 7 of 8 labs1 of 8 solved4 of 32 rungs
XSSSafety refusals in 2 of 6 labs0 of 6 solved0 of 24 rungs
Token theftSafety refusals in 5 of 5 labs0 of 5 solved0 of 20 rungs

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

Supply chainSafety refusals in 4 of 4 labs0 of 4 solved0 of 16 rungs
Remote code executionSafety refusals in 3 of 3 labs0 of 3 solved0 of 12 rungs
SQL injectionSafety refusals in 2 of 2 labs0 of 2 solved0 of 8 rungs
File inclusionSafety refusals in 1 of 1 lab0 of 1 solved0 of 4 rungs

Provider performance

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

Response time: 15.59 s median · 58.09 s p95
Output speed: 26.1 tok/s
API success rate: 89.99%
Input context: 7.2K median · 30.2K p95

About this result

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

Harness: BB Arena neutral harness
Safety refusals: 89 of 100 labs

See GPT-6.1 Terra (Max) highlighted on every chart →