SEC-Bench Pro
We show this table for reference; we do not rank on it.
Cybersecurity benchmark for agentic vulnerability analysis and exploit-oriented security tasks.
Benchmark score on SEC-Bench Pro — November 13, 2026
We compile the SEC-Bench Pro rows from provider self-reports. Gemini 4 Carbon leads the table at 88.2%, followed by GPT-6 Cyber (87.0%) and GPT-6 Astra (85.4%). We do not use these results to rank models overall.
Gemini 4 Carbon
GPT-6 Cyber
OpenAI
Reported by OpenAIGPT-6 Astra
OpenAI
Reported by OpenAI13 modelsAgenticCurrentDisplay onlyUpdated November 13, 2026
Benchmark score table (13 models)
All 13 results on this page were reported by the labs that built the models.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Gemini 4 CarbonGoogleReported by Google | 88.2% | Reported by Google | Not reported | Closed |
| 2 | GPT-6 CyberOpenAIReported by OpenAI | 87.0% | Reported by OpenAI | Not reported | Closed |
| 3 | GPT-6 AstraOpenAIReported by OpenAI | 85.4% | Reported by OpenAI | Not reported | Closed |
| 4 | Gemini 4 ArgonGoogleReported by Google | 83.4% | Reported by Google | Not reported | Closed |
| 5 | GPT-6.1 TerraOpenAIReported by OpenAI | 83.3% | Reported by OpenAI | Not reported | Closed |
| 6 | GPT-6.1 SolOpenAIReported by OpenAI | 78.8% | Reported by OpenAI | Not reported | Closed |
| 7 | GPT-5.6 SolOpenAIReported by OpenAI | 71.2% | Reported by OpenAI | Not reported | Closed |
| 8 | GPT-6 SolOpenAIReported by OpenAI | 66.3% | Reported by OpenAI | Not reported | Closed |
| 9 | MiMo-V2.6-ProXiaomiReported by Xiaomi | 66.3% | Reported by Xiaomi | Not reported | Open |
| 10 | DeepSeek V4.1 Flash 1017DeepSeekReported by DeepSeek | 62.8% | Reported by DeepSeek | Not reported | Open |
| 11 | MiMo-V2.6-FlashXiaomiReported by Xiaomi | 47.5% | Reported by Xiaomi | Not reported | Open |
| 12 | GPT-5.5OpenAIReported by OpenAI | 45.8% | Reported by OpenAI | Not reported | Closed |
| 13 | GPT-6 LunaOpenAIReported by OpenAI | 34.2% | Reported by OpenAI | Not reported | Closed |
- Lab-reported
- Benchmark-run
- Independent
Among the reported SEC-Bench Pro rows, Gemini 4 Carbon is first at 88.2%. The third row is 2.8 points behind. The broader top-10 range is 25.4 points, so the table still separates the published systems.
13 models have been evaluated on SEC-Bench Pro. The benchmark falls in the Agentic category. We keep these results separate from the weighted global scoring system, so they remain source-specific evidence. SEC-Bench Pro is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About SEC-Bench Pro
Year
2026
Tasks
Security engineering tasks
Format
Success rate
Difficulty
Advanced cybersecurity
OpenAI reports exact model results on SEC-Bench Pro in its GPT-5.6 launch table. BenchLM stores the provider-run values as display-only cyber evidence until benchmark-native results are available.
SEC-Bench Pro is one of the cyber benchmarks we track. The cybersecurity benchmark comparison sets it beside the others, one table per security job.
Freshness and provenance
Version
SEC-Bench Pro 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does SEC-Bench Pro measure?
Cybersecurity benchmark for agentic vulnerability analysis and exploit-oriented security tasks.
Which model scores highest on SEC-Bench Pro?
Gemini 4 Carbon by Google currently leads with a score of 88.2% on SEC-Bench Pro.
How many models are evaluated on SEC-Bench Pro?
13 AI models have published results on SEC-Bench Pro in the BenchLM catalog.
Compare top models on SEC-Bench Pro
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.