✦ Our pick · Security & vulnerabilities
GPT-6 Sol at high effort, $16.32 per finished task, is the cheapest you can get today.
Updated Oct 13, 2026, every 3 hours. See it live with your plans and costs.
The top 10 for security & vulnerabilities
Find and fix vulnerabilities without breaking the code. Cost per finished task: the tokens at each model's best effort level for this work, plus your time fixing failed attempts ($93 an hour). Success is the share of such tasks a model gets right the first time, on the benchmarks below.
| # | Model | Effort | Success | $/task |
|---|---|---|---|---|
| 1 | GPT-6 Sol OpenAI | high | 69.0% measured | $16.32 |
| 2 | Gemini 4 Argon Google | high | 67.2% partly measured | $17.63 |
| 3 | MiMo-V2.6-Pro Xiaomi | default | 61.4% measured | $19.73 |
| 4 | GPT-5.6 Sol OpenAI | medium | 64.5% partly measured | $20.45 |
| 5 | DeepSeek V4.1 Flash DeepSeek | max | 60.3% measured | $20.75 |
| 6 | MiMo-V2.6-Flash Xiaomi | default | 59.9% partly measured | $20.86 |
| 7 | Claude Haiku 5.5 Anthropic | max | 60.2% partly measured | $20.95 |
| 8 | GPT-6 Luna OpenAI | max | 59.8% measured | $21.08 |
| 9 | Grok 4.7 SpaceXAI | xhigh | 62.0% measured | $21.53 |
| 10 | GPT-5.6 Luna OpenAI | max | 59.1% partly measured | $22.11 |
Best within your tool
| Tool | Model | Effort | $/task |
|---|---|---|---|
| Cursor | Claude Haiku 5.5 | max | $20.95 |
| Claude Code | Claude Haiku 5.5 | max | $20.95 |
| Codex | GPT-6 Sol | high | $16.32 |
| Antigravity | Claude Sonnet 5.5 | xhigh | $23.56 |
| GitHub Copilot | GPT-6 Sol | high | $16.32 |
If tokens come with your plan
Tokens included in a plan cost you nothing extra, so the pick is the most reliable model in it, at the leanest effort level within our error of the best (the top levels only use up the plan sooner).
| Plan | Model | Effort |
|---|---|---|
| Claude Pro | Claude Haiku 5.5 | max |
| Claude Max | Claude Haiku 5.5 | max |
| ChatGPT Plus or Business | GPT-6 Sol | xhigh |
| ChatGPT Pro | GPT-6 Sol | xhigh |
| Google AI Pro | Claude Sonnet 5.5 | xhigh |
| Google AI Ultra | Claude Sonnet 5.5 | xhigh |
Measured on
- CWE-Bench-AA · Artificial Analysis. 120 held-out tasks: find and fix OWASP Top-10 vulnerabilities in a repository without breaking its behaviour (pass@1). Measured for 15 of the 42 ranked models, results as of Oct 13, 2026.
- Vals Cyber · Vals AI. Craft inputs that trigger real OSS-Fuzz vulnerabilities (mostly C/C++ memory safety), then patch the source so they no longer do. Measured for 30 of the 42 ranked models, results as of Oct 12, 2026.
- CyberGym-E2E-AA · Artificial Analysis. Find real memory-safety vulnerabilities in C/C++ open-source projects and prove them with a working input (Berkeley RDI's CyberGym-E2E, run by Artificial Analysis). Measured for 15 of the 42 ranked models, results as of Oct 13, 2026.
- DeepsecBench-AA · Artificial Analysis. Find vulnerabilities in real open-source application code (Vercel's DeepsecBench, run by Artificial Analysis). Measured for 16 of the 42 ranked models, results as of Oct 13, 2026.
- Reverse engineering (SRE Bench) · Vals AI. Reverse engineer real, contamination-free binaries: work out what compiled programs do. Measured for 26 of the 42 ranked models, results as of Oct 8, 2026.
How every number is worked out, with the errors of our estimates: the method.