DeepSec-Bench (AA)
Deeper security engineering tasks than a single patch: reason about a system and defend it. Run by Artificial Analysis.
As of 18 Oct 2026, GPT-6 Sol leads DeepSec-Bench (AA) on BenchLeader with 46.3%, ahead of GPT-6 Astra at 36.9%, across 19 model configurations with a published result.
- Published by
- Artificial Analysis
- Category
- Agents & tools
- Index weight
- 0.5
- Models
- 19
- Data as of
- 18 Oct 2026
Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.
- 1GPT-6 Sol (max)46.3%
- 2GPT-6 Astra (max)36.9%
- 3DeepSeek V4.1 Flash (max)31.2%
- 4Grok 4.7 (xhigh)26.9%
- 5Claude Opus 5.5 (max)26.9%
- 6MiMo-V2.6-Pro26.2%
- 7Claude Opus 5 (max)25.9%
- 8GLM 5.3 (max)24.3%
- 9GPT-6 Luna (max)23.5%
- 10Claude Fable 5.1 (max)23.5%
- 11GLM 5.3 Flash19.9%
- 12Mistral Large 416.0%
- 13Step 5 Preview15.8%
- 14Gemini 3.8 Flash (high)14.5%
- 15Kimi K3 (max)12.4%
19 of 19
| # | ||||
|---|---|---|---|---|
| 1 |  GPT-6 Solmax | 46.3% | 65.6 | 2026-09-22 |
| 2 | GPT-6 Astramax | 36.9% | 69.7 | 2026-09-03 |
| 3 |  DeepSeek V4.1 Flashmaxopen ↗ | 31.2% | 61.5 | 2026-09-10 |
| 4 |  Grok 4.7xhigh | 26.9% | 60.6 | 2026-09-21 |
| 5 |  Claude Opus 5.5max | 26.9% | 71.6 | 2026-09-22 |
| 6 |  MiMo-V2.6-Proopen ↗ | 26.2% | 65.3 | 2026-09-21 |
| 7 | Claude Opus 5max | 25.9% | 67.9 | 2026-07-24 |
| 8 |  GLM 5.3maxopen ↗ | 24.3% | 63.6 | 2026-08-18 |
| 9 | GPT-6 Lunamax | 23.5% | 58.6 | 2026-09-22 |
| 10 | Claude Fable 5.1max | 23.5% | 69.9 | 2026-09-01 |
| 11 | GLM 5.3 Flashopen ↗ | 19.9% | 62.2 | 2026-08-26 |
| 12 |  Mistral Large 4open | 16.0% | 58.7 | 2026-10-06 |
| 13 |  Step 5 Previewopen | 15.8% | 62.2 | 2026-09-18 |
| 14 |  Gemini 3.8 Flashhigh | 14.5% | 63.3 | 2026-09-02 |
| 15 |  Kimi K3maxopen ↗ | 12.4% | 64.8 | 2026-07-16 |
| 16 |  Qwen3.8 27Bxhighopen ↗ | 11.5% | 56.3 | 2026-08-14 |
| 17 | New Beam 1.0open | 6.9% | 56.4 | 2026-10-16 |
| 18 |  MiniMax M3open ↗ | 1.1% | 56.2 | 2026-06-01 |
| 19 |  Nemotron 3 Ultra 550B A55Bthinkingopen ↗ | 1.1% | 55.6 | 2026-06-04 |
More agents & tools benchmarks · All LLM benchmarks
Cite as: BenchLeader, “DeepSec-Bench (AA) leaderboard”, https://www.benchleader.com/benchmarks/aa_deepsec_bench, data as of 18 Oct 2026.
DeepSec-Bench (AA): questions
- What does DeepSec-Bench (AA) measure?
- Deeper security engineering tasks than a single patch: reason about a system and defend it. Run by Artificial Analysis. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads DeepSec-Bench (AA)?
- GPT-6 Sol leads DeepSec-Bench (AA) with 46.3% as of 18 Oct 2026, ahead of GPT-6 Astra at 36.9%.
- How many models have DeepSec-Bench (AA) results?
- 19 model configurations have a DeepSec-Bench (AA) result on BenchLeader, all taken from Artificial Analysis.
- Who runs DeepSec-Bench (AA) and how often is it updated?
- DeepSec-Bench (AA) is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does DeepSec-Bench (AA) count toward the BenchLeader Index?
- Yes. DeepSec-Bench (AA) contributes to the agents & tools category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.