OpenAI · Proprietary · tested 8 Nov 2026
GPT-6 Astra High effort
Safety refusals affected this result
OpenAI's cyber safeguards stopped attempts in 76 of 100 labs. Stopped attempts score zero. BB Arena is enrolled in OpenAI's cyber program for verified security researchers, but its tier doesn't cover GPT-6 Astra.
With refusals in more than 10 of 100 labs, its score says more about the safeguards than the model, so it isn't ranked: it's on the refusals board. Its cost, tokens and time are left out of comparisons too. How refusals are scored
On the labs it was never refused
OpenAI's safeguards stopped attempts in 76 of 100 labs. On the other 24, GPT-6 Astra worked without refusals, so here it's compared like for like with the ranked models that did best on those same 24 labs. These labs aren't a random sample of the benchmark, so this doesn't predict a full score.
- GPT-6 Astra (High)not ranked0 of 24 solved0 of 96 rungs
- Grok 4.6 (High)#2 overall8 of 24 solved53 of 96 rungs
- Grok 4.7 (High)#4 overall6 of 24 solved44 of 96 rungs
- Claude Opus 5 (High)#1 overall6 of 24 solved42 of 96 rungs
- DeepSeek V4.1 Flash (Max)#3 overall5 of 24 solved41 of 96 rungs
By vulnerability class
How GPT-6 Astra (High) did on each type of vulnerability. Shown as counts: each class has 10 labs, so one lab is 10% of a class.
- Access controlSafety refusals in 8 of 10 labs0 of 10 solved0 of 40 rungs
- Authentication & sessionsSafety refusals in 7 of 10 labs0 of 10 solved0 of 40 rungs
- Injection: queries & dataSafety refusals in 9 of 10 labs0 of 10 solved0 of 40 rungs
- Code executionSafety refusals in 4 of 10 labs0 of 10 solved0 of 40 rungs
- Server-side accessSafety refusals in 8 of 10 labs0 of 10 solved0 of 40 rungs
- Browser-sideSafety refusals in 8 of 10 labs0 of 10 solved0 of 40 rungs
- Infrastructure & edgeSafety refusals in 9 of 10 labs0 of 10 solved0 of 40 rungs
- Business logic & racesSafety refusals in 8 of 10 labs0 of 10 solved0 of 40 rungs
- Secrets, crypto & supply chainSafety refusals in 6 of 10 labs0 of 10 solved0 of 40 rungs
- AI agentsSafety refusals in 9 of 10 labs0 of 10 solved0 of 40 rungs
Provider performance
Measured per single agent against the provider endpoint used for the run.
- Response time
- 8.51 s median · 16.66 s p95
- Output speed
- 13.2 tok/s
- API success rate
- 84.02%
- Input context
- 2.2K median · 7.2K 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
- 76 of 100 labs