# Cybersecurity leaderboard

As of November 21, 2026, Gemini 4 Carbon leads our cybersecurity leaderboard at 90.0 (Estimated, $2/$10 per 1M tokens), ahead of Claude Fable 5.2 at 87.0. GPT-5.5 is the highest Supported row at #8 (64.7); Qwen3.8 Max leads open-weight models at #10 (61.6). We rank 47 models (13 Supported, 34 Estimated) across 7 domain protocols, and this vertical carries zero weight in the overall score.

## Shortlist by workload and budget

- **Highest point score:** [Gemini 4 Carbon](/models/gemini-4-carbon) (#1, 90.0, Estimated, $2/$10 per 1M tokens) — Carries an Estimated label; pick GPT-5.5 (#8, 64.7) when you require multi-source Supported coverage.
- **Highest Supported row:** [GPT-5.5](/models/gpt-5-5) (#8, 64.7, Supported, $5/$30 per 1M tokens) — Trails Gemini 4 Carbon by 25.3 points, so you trade peak point score for broader independent verification.
- **Top open-weight model:** [Qwen3.8 Max](/models/qwen3-8-max) (#10, 61.6, Estimated, Open weights · self-hostable) — Trails Gemini 4 Carbon by 28.4 points (#10 overall here), so pick it when weight control or on-prem residency outweighs peak score.

## Ranked models

| Rank | Model | Creator | Score | Conditional range | Evidence | API ($/1M in/out) | Context | Source families |
|---:|---|---|---:|---:|---|---:|---:|---:|
| #1 | [Gemini 4 Carbon](/models/gemini-4-carbon) | Google | 90.0 | 73.8-100.0 | Estimated | $2.00 / $10.00 | 1M | 3 |
| #2 | [Claude Fable 5.2](/models/claude-fable-5-2) | Anthropic | 87.0 | 67.9-100.0 | Estimated | $8.00 / $40.00 | 1M | 1 |
| #3 | [GPT-6 Astra](/models/gpt-6-astra) | OpenAI | 82.0 | 68.6-95.4 | Estimated | $10.00 / $50.00 | 1.05M | 4 |
| #4 | [Gemini 4 Argon](/models/gemini-4-argon) | Google | 79.3 | 65.8-92.8 | Estimated | $2.00 / $10.00 | 1M | 3 |
| #5 | [Claude Opus 5](/models/claude-opus-5) | Anthropic | 73.4 | 55.5-91.3 | Estimated | $5.00 / $25.00 | - | 2 |
| #6 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | OpenAI | 70.3 | 57.8-82.8 | Estimated | $4.00 / $20.00 | 1.05M | 5 |
| #7 | [Muse Wonder](/models/muse-wonder) | Meta | 68.8 | 50.1-87.6 | Estimated | $4.00 / $20.00 | 1M | 1 |
| #8 | [GPT-5.5](/models/gpt-5-5) | OpenAI | 64.7 | 52.0-77.5 | Supported | $5.00 / $30.00 | 1M | 3 |
| #9 | [Inkling 2 Small](/models/inkling-2-small) | Thinking Machines Lab | 63.9 | 44.1-83.7 | Estimated | $0.30 / $1.00 | 1M | 1 |
| #10 | [Qwen3.8 Max](/models/qwen3-8-max) | Alibaba | 61.6 | 43.3-79.9 | Estimated | - | 1M | 1 |
| #11 | [Kimi K3](/models/kimi-k3) | Moonshot AI | 59.4 | 40.6-78.3 | Supported | $3.00 / $15.00 | 1.05M | 3 |
| #12 | [GPT-5.4](/models/gpt-5-4) | OpenAI | 59.4 | 46.3-72.5 | Supported | $2.50 / $15.00 | 1.05M | 3 |
| #13 | [Kimi K3.1 Pro](/models/kimi-3-1-pro) | Moonshot AI | 61.3 | 41.6-81.0 | Estimated | $2.00 / $10.00 | 1.05M | 1 |
| #14 | [GLM-5.3](/models/glm-5-3) | Z.AI | 58.8 | 41.1-76.6 | Estimated | Free | 1M | 2 |
| #15 | [Gemini 3.6 Flash](/models/gemini-3-6-flash) | Google | 60.8 | 42.2-79.5 | Estimated | $0.75 / $3.75 | 1M | 1 |
| #16 | [Gemini 3.1 Pro](/models/gemini-3-1-pro) | Google | 58.7 | 41.0-76.4 | Supported | $2.00 / $12.00 | 1M | 2 |
| #17 | [Claude Opus 4.7](/models/claude-opus-4-7) | Anthropic | 58.7 | 44.9-72.4 | Estimated | $5.00 / $25.00 | 1M | 2 |
| #18 | [Gemini 3 Pro](/models/gemini-3-pro) | Google | 58.2 | 38.5-77.8 | Estimated | $2.00 / $12.00 | 2M | 1 |
| #19 | [Claude Opus 4.6](/models/claude-opus-4-6) | Anthropic | 56.2 | 40.4-71.9 | Supported | $5.00 / $25.00 | 1M | 2 |
| #20 | [Claude Opus 4.6 (Adaptive)](/models/claude-opus-4-6-thinking) | Anthropic | 56.2 | 40.4-71.9 | Supported | - | 1M | 2 |
| #21 | [Qwen3.7 Max](/models/qwen3-7-max) | Alibaba | 56.2 | 38.0-74.5 | Estimated | - | 1M | 1 |
| #22 | [GPT-5.3 Codex](/models/gpt-5-3-codex) | OpenAI | 56.0 | 34.6-77.5 | Estimated | $1.75 / $14.00 | 400K | 1 |
| #23 | [GLM-5.2](/models/glm-5-2) | Z.AI | 55.4 | 37.1-73.7 | Estimated | $1.40 / $4.40 | 1M | 1 |
| #24 | [Claude Sonnet 4.6](/models/claude-sonnet-4-6) | Anthropic | 52.5 | 37.3-67.7 | Supported | $3.00 / $15.00 | 200K | 2 |
| #25 | [DeepSeek V4 Pro](/models/deepseek-v4-pro) | DeepSeek | 54.4 | 35.7-73.2 | Estimated | $1.32 / $3.96 | 1M | 1 |
| #26 | [Grok 4.5](/models/grok-4-5) | xAI | 53.9 | 33.1-74.8 | Estimated | $2.00 / $6.00 | 500K | 1 |
| #27 | [GPT-5.2](/models/gpt-5-2) | OpenAI | 53.7 | 33.5-73.8 | Estimated | $1.75 / $14.00 | 400K | 1 |
| #28 | [GLM-5.1](/models/glm-5-1) | Z.AI | 50.7 | 34.7-66.7 | Supported | $1.40 / $4.40 | 203K | 2 |
| #29 | [Muse Spark 1.1](/models/muse-spark-1-1) | Meta | 50.5 | 37.8-63.2 | Estimated | - | 1M | 3 |
| #30 | [Claude Opus 4.5](/models/claude-opus-4-5) | Anthropic | 48.1 | 31.7-64.6 | Supported | $5.00 / $25.00 | 200K | 2 |
| #31 | [Claude Opus 4.5 Thinking](/models/claude-opus-4-5-thinking) | Anthropic | 48.1 | 31.7-64.6 | Supported | - | 200K | 2 |
| #32 | [Gemini 3 Flash](/models/gemini-3-flash) | Google | 49.9 | 29.9-69.8 | Estimated | $0.50 / $3.00 | 1M | 1 |
| #33 | [GPT-5.2-Codex](/models/gpt-5-2-codex) | OpenAI | 49.7 | 29.2-70.2 | Estimated | $1.75 / $14.00 | 400K | 1 |
| #34 | [Inkling](/models/inkling) | Thinking Machines Lab | 48.8 | 30.1-67.5 | Estimated | $1.87 / $4.68 | 1M | 1 |
| #35 | [Grok 4](/models/grok-4) | xAI | 45.6 | 24.2-67.0 | Estimated | - | 128K | 1 |
| #36 | [Claude Sonnet 4.5](/models/claude-sonnet-4-5) | Anthropic | 44.3 | 24.9-63.7 | Estimated | $3.00 / $15.00 | 200K | 1 |
| #37 | [Claude Sonnet 4.5 Thinking](/models/claude-sonnet-4-5-thinking) | Anthropic | 44.3 | 24.9-63.7 | Estimated | - | 200K | 1 |
| #38 | [Kimi K2.5](/models/kimi-k2-5) | Moonshot AI | 42.9 | 23.4-62.4 | Estimated | $0.60 / $3.00 | 256K | 1 |
| #39 | [GLM-4.7](/models/glm-4-7) | Z.AI | 41.4 | 21.2-61.6 | Estimated | Free | 200K | 1 |
| #40 | [Gemini 2.5 Pro](/models/gemini-2-5-pro) | Google | 41.3 | 21.2-61.5 | Estimated | $1.25 / $10.00 | 1M | 1 |
| #41 | [Beam 1.0 501B-A23B](/models/reflection-beam) | Reflection AI | 40.2 | 15.8-64.6 | Estimated | - | 1M | 1 |
| #42 | [GLM-5](/models/glm-5) | Z.AI | 37.6 | 20.7-54.4 | Supported | $1.00 / $3.20 | 200K | 2 |
| #43 | [Grok 4.1 Fast](/models/grok-4-1-fast) | xAI | 37.7 | 14.8-60.5 | Estimated | $0.20 / $0.50 | 1M | 1 |
| #44 | [Muse Spark](/models/muse-spark) | Meta | 37.0 | 19.7-54.2 | Estimated | - | 262K | 1 |
| #45 | [Claude Haiku 4.5](/models/claude-haiku-4-5) | Anthropic | 33.4 | 14.6-52.2 | Supported | $1.00 / $5.00 | 200K | 2 |
| #46 | [Claude Haiku 4.5 Thinking](/models/claude-haiku-4-5-thinking) | Anthropic | 33.4 | 14.6-52.2 | Supported | - | 200K | 2 |
| #47 | [DeepSeek V3.2](/models/deepseek-v3-2) | DeepSeek | 35.2 | 14.3-56.1 | Estimated | $0.28 / $0.42 | 128K | 1 |

## Scored domain benchmarks (7)

- [CyberGym](/benchmarks/cybergym) (32% ref. weight): A cybersecurity task benchmark for evaluating defensive cyber workflows and vulnerability-oriented agent performance.
- [ExploitGym](/benchmarks/exploitgym) (27% ref. weight): A controlled benchmark for evaluating whether AI agents can extend vulnerability-triggering inputs into working exploits.
- PromptArmor Frontier Breach · PromptArmor (13% ref. weight): Enterprise indirect prompt-injection defense across untrusted documents, emails, and tool outputs.
- [ExploitBench (official board)](/benchmarks/exploitbench) (10% ref. weight): A cybersecurity benchmark for evaluating LLM agents on full-control V8 exploit synthesis using 16 measured exploit capability flags.
- [BinaryAudit](/benchmarks/binaryaudit) (10% ref. weight): Quesma's BinaryAudit asks AI agents to find backdoors and timebombs planted in compiled, stripped binaries of real server software, using the reverse engineering tools Ghidra and radare2. The score here is the share of detection runs that found the planted code.
- [ExploitBench](/benchmarks/exploitbench) (5% ref. weight): A cybersecurity benchmark for evaluating LLM agents on full-control V8 exploit synthesis using 16 measured exploit capability flags.
- [Gray Swan IPI (15 attempts)](/benchmarks/gray-swan-ipi) (5% ref. weight): Measures how often indirect prompt-injection attacks succeed within fifteen attempts. Lower scores mean greater resistance to the tested attacks.

## Patch reproduction, exploit synthesis, and prompt-injection defense sit on one calibrated scale

This board combines seven open and benchmark-owner security protocols led by CyberGym (32%), ExploitGym (27%), PromptArmor Frontier Breach (13%). CyberGym and BinaryAudit test whether an agent can locate and reproduce real software vulnerabilities inside containerized targets, ExploitGym and ExploitBench test whether it can turn crash inputs into working exploits, and PromptArmor Frontier Breach plus Gray Swan IPI measure whether untrusted inputs hijack the model.

We calibrate each protocol with BenchAlign v6.0 item-response difficulty parameters instead of averaging raw percentages, because a 30% solve rate on containerized exploit tasks is harder than an 85% defense rate on prompt-injection suites. Every scored row must also qualify for the general catalog, which keeps narrow single-task harnesses from distorting the scale.

## Why we isolate cybersecurity from the overall score and exclude restricted cyber profiles

Dual-use offensive capability never moves our overall ranking: a model does not climb the main leaderboard by writing better V8 exploits. Vendor-gated cyber profiles such as GPT-6 Cyber under OpenAI Daybreak Red and Claude Mythos under Anthropic's verification program are also left off this general-access table because most engineering teams cannot call them over standard API keys.

A high score here does not clear a model for autonomous SOC triage, production firewall rules, or unreviewed code patching, and it never authorizes offensive testing. When you need per-benchmark numbers including restricted-access models, use the task tables on /cybersecurity rather than a composite score.

[Inspect restricted cyber models & per-test tables](/cybersecurity) · [Gemini 4 Carbon vs Claude Fable 5.2](/compare/claude-fable-5-2-vs-gemini-4-carbon)

## Choosing between frontier APIs and self-hosted security models

Gemini 4 Carbon (90.0, $2/$10 per 1M tokens) leads Claude Fable 5.2 (87.0, $8/$40 per 1M tokens) by 3.0 points, while GPT-5.5 (#8, 64.7) is the highest row with Supported multi-evaluator coverage. Because Gemini 4 Carbon already prices at $2/$10 per 1M tokens, you do not have to pay flagship $8/$40 token rates to get the top point estimate on this board.

Teams auditing proprietary repositories that cannot leave their own VPC should start with Qwen3.8 Max (#10, 61.6), then budget for the 28.4-point capability gap against Gemini 4 Carbon by pairing the model with deterministic static-analysis scanners and unit-test gates.

[Compare local & self-hosted LLMs](/best/local-llm) · [Check API price vs performance](/llm-price-performance)

## Questions

### Which LLM is best for cybersecurity right now?

Gemini 4 Carbon holds #1 on our cybersecurity leaderboard at 90.0 (Estimated) as of November 21, 2026, ahead of Claude Fable 5.2 at 87.0 and GPT-6 Astra at 82.0. GPT-5.5 (#8, 64.7) is the highest Supported option when independent multi-evaluator depth matters most.

### What is the best open-weight LLM for cybersecurity?

Qwen3.8 Max from Alibaba leads open-weight models on this cybersecurity board at #10 with a calibrated score of 61.6 (Estimated). That trails Gemini 4 Carbon by 28.4 points, so choose Qwen3.8 Max when self-hosting, VPC data residency, or weight customization matters more than peak frontier accuracy.

### Which API model offers the best value for cybersecurity?

Gemini 4 Carbon leads both point accuracy (90.0) and price efficiency here at $2/$10 per 1M tokens, undercutting $8/$40 flagship tiers without sacrificing cybersecurity quality. If you need an even lower input rate for high-volume batch runs, compare mid-tier alternatives in the table below against your own retry rate.

### Which benchmarks make up the cybersecurity leaderboard score?

BenchAlign v6.0 calibrates 7 open and primary domain protocols for this board, led by CyberGym (32%), ExploitGym (27%), PromptArmor Frontier Breach (13%), ExploitBench (official board) (10%). Each percentage reflects share of the vertical's reference weight inside our item-response model rather than a raw score average, and no single independent evaluator can exceed 40% when three or more evaluators contribute.

### Why are GPT-6 Cyber and Claude Mythos absent from this cybersecurity leaderboard?

This leaderboard only ranks general-access models eligible for the main catalog, so engineering teams can deploy what they see. Vendor-gated cyber profiles such as GPT-6 Cyber (Daybreak Red) and Claude Mythos require program approval and run under custom lab harnesses; we track their per-benchmark results separately on /cybersecurity.

### Do cybersecurity benchmark scores affect the overall LLM ranking?

No. Cybersecurity is an isolated specialist vertical with zero weight in our overall score, ensuring dual-use exploit generation never boosts a model's general rank. Among the 47 models here, 13 hold Supported labels and 34 hold Estimated labels, with single-source estimates held behind close Verified contenders.

## Related benchmarks and guides

- [Cybersecurity Benchmarks by Task](/cybersecurity): Per-benchmark tables, restricted Daybreak and Mythos rows, and vendor access rules.
- [Coding Leaderboard](/coding): Repository bug fixing and software engineering rankings on SWE-bench Pro and LiveCodeBench.
- [Agentic Leaderboard](/agentic): Terminal execution, tool orchestration, and autonomous workflow rankings.
- [Best Local LLMs](/best/local-llm): Self-hosted open-weight models for air-gapped security review.

## Other specialist vertical leaderboards

- [Documents and extraction leaderboard](/leaderboards/documents) — #1 GPT-6.2 Astra (86.5)
- [Factuality and anti-hallucination leaderboard](/leaderboards/factuality) — #1 Gemini 4 Carbon (78.8)
- [Finance and accounting leaderboard](/leaderboards/finance) — #1 Gemini 4 Carbon (89.2)
- [Frontend and web development leaderboard](/leaderboards/frontend) — #1 Claude Fable 5.2 (89.5)
- [Healthcare and medical leaderboard](/leaderboards/healthcare) — #1 Claude Fable 5.2 (87.5)
- [Legal leaderboard](/leaderboards/legal) — #1 Claude Fable 5.2 (86.1)
- [Science and research leaderboard](/leaderboards/science) — #1 Claude Fable 5.2 (94.1)
- [Search and deep research leaderboard](/leaderboards/search) — #1 Claude Fable 5.2 (86.6)

Last updated: November 21, 2026

Canonical page: https://benchlm.ai/leaderboards/cybersecurity
