← Leaderboard
Anthropic · Proprietary · tested 6 Oct 2026
Claude Opus 5 High effort
Overall rank#2 of 11
Solves60/10053 first attempt · +7 retry
Rungs proven293/400267 first attempt · +26 retry
Cost to run$463$7.72 per solve
Tokens used312.5M92.0% from cache · 8.7M output
Time per lab17.1 minaverage of 147 attempts
Where it places
Each line is one metric across all 11 configurations. Right is always better; this model is the large dot.
Solveslabs solved, of 100
worsebetter
60#2 of 11
Rungs provenrungs proven, of 400
worsebetter
293#3 of 11
First-attempt solvessolved on the first attempt
worsebetter
53#3 of 11
Cost per solveUSD per solved lab
worsebetter
$7.72#7 of 7
By vulnerability class
How Claude Opus 5 (High) did on each type of vulnerability. Shown as counts, because some classes have only a few labs.
- Authentication / authorisation bypass17 of 28 solved85 of 112 rungs
- Information disclosure10 of 12 solved43 of 48 rungs
- IDOR3 of 11 solved25 of 44 rungs
- SSRF2 of 10 solved16 of 40 rungs
- Business logic5 of 10 solved29 of 40 rungs
- Key or secret exposure8 of 8 solved32 of 32 rungs
- XSS4 of 6 solved17 of 24 rungs
- Token theft5 of 5 solved20 of 20 rungs
Fewer than 5 labs each: one lab changes these a lot, so read them with care.
- Supply chain4 of 4 solved16 of 16 rungs
- Remote code execution0 of 3 solved1 of 12 rungs
- SQL injection2 of 2 solved8 of 8 rungs
- File inclusion0 of 1 solved1 of 4 rungs
Provider performance
Measured per single agent against the provider endpoint used for the run.
- Response time
- 11.50 s median · 37.97 s p95
- Output speed
- 65.7 tok/s
- API success rate
- 99.08%
- Input context
- 31.6K median · 74.4K 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
- None