Who is SOTA in Cybersecurity?Crown open.
Agentic security work on private sets from Vals AI. CyberBench (reproduce and patch OSS-Fuzz vulnerabilities) is the primary; SRE Bench (reverse engineering) ranks the rest. Source: Vals AI. As of Nov 3, 2026, Cybersecurity is uncrowned on isitsota (methodology v1.0). Crown history, benchmark weights and methodology.
| # | Model | State | Confidence |
|---|---|---|---|
| 1 | ZGLM 5.3 Zhipu AI | Near-SOTA · Cybersecurity | medium |
| 2 | XIMimo V2.6 Flash Xiaomi | Near-SOTA · Cybersecurity | medium |
| 3 | XIMimo V2.6 Pro Xiaomi | Near-SOTA · Cybersecurity | medium |
| 2 | KKimi K3 Moonshot AI | #2 in Cybersecurity | low |
| 4 | TEHy4 Preview Tencent | #4 in Cybersecurity | medium |
| 5 | DSDeepSeek V4.1 Flash DeepSeek | #5 in Cybersecurity | medium |
| 5 | ZGLM 5.3 Flash Zhipu AI | #5 in Cybersecurity | medium |
| 6 | ZGLM 5.2 Zhipu AI | #6 in Cybersecurity | medium |
| 6 | THInkling thinky | #6 in Cybersecurity | medium |
| 7 | QQwen3.8 27B Alibaba / Qwen | #7 in Cybersecurity | low |
| 10 | MIMinimax M3 MiniMax | #10 in Cybersecurity | low |
| 11 | THInkling Small thinky | #11 in Cybersecurity | low |
Best page and recent changes
Best AI model for cybersecurity
- Cybersecurity (best per dollar): Claude Haiku 5.5 and Gemini 3.8 Flash and GLM 5.3 and GLM 5.3 Flash and GPT-6.1 Sol and GPT-6 Astra and Mimo V2.6 Flash are co-SOTA
- Cybersecurity (best per dollar): Claude Haiku 5.5 and DeepSeek V4.1 Flash and Gemini 3.8 Flash and GLM 5.3 and GPT-6.1 Sol and GPT-6 Astra and Mimo V2.6 Flash are co-SOTA
- Cybersecurity (best per dollar): Claude Haiku 5.5 and DeepSeek V4.1 Flash and Gemini 3.8 Flash and GLM 5.3 and GPT-6.1 Sol and GPT-6 Astra and Mimo V2.6 Flash are co-SOTA
What can move this crown
- CyberBench — primary, weight 1 · Source: Vals AI · Automatic (tests or exact checks)
- SRE Bench — secondary, weight 0.5 · Source: Vals AI · Automatic (tests or exact checks)
Source: Vals AI for the Vals benchmarks above: proprietary private test sets run by Vals and republished with permission. Only each benchmark's standard harness counts; product-harness runs (Claude Code, Codex and similar) are not matched to base models. Vals is the only source on this board, so after the first holder a crown change needs two consecutive Vals snapshots that agree. A snapshot is a new Vals release of the board, dated the day isitsota first saw it.
Who is SOTA in Cybersecurity?
As of Nov 3, 2026, Cybersecurity is uncrowned on isitsota (methodology v1.0).
Since when is the Cybersecurity crown current?
As of Nov 3, 2026, Cybersecurity has an open crown.