dfbench Detection Recall
Cybersecurity · 2026-08-19
depthfirst's held-out defensive security benchmark, built from real vulnerabilities in open-source projects. An agent audits a repository scope in depthfirst's internal harness with network access disabled, and detection recall is the share of the scope's known vulnerabilities it finds. The August 19, 2026 post describes 253 examples and 910 vulnerabilities, and the page says the dataset will keep evolving. depthfirst ran these evaluations under providers' trusted access programs without saying which rows did, so scores aren't established as the stock models' and stay diagnostic.
Benchmark source: Depthfirst
Top models (higher is better)
| Model | Score | Source |
|---|---|---|
| GPT-5.6 Sol | 65.7 | dfbench: GPT 5.6 Sol xhigh · depthfirst |
| GPT-6 Sol | 59.3 | dfbench: GPT 6 Sol xhigh · depthfirst |
| Grok 4.7 | 59.0 | dfbench: Grok 4.7 xhigh · depthfirst |
| DeepSeek V4.1 Flash | 57.3 | dfbench: DeepSeek v4.1 Flash high · depthfirst |
| Grok 4.5 | 57.2 | dfbench: Grok 4.5 high · depthfirst |
| MiMo-V2.6-Pro | 55.0 | dfbench: MiMo-v2.6 Pro · depthfirst |
| Claude Opus 5.5 | 54.9 | dfbench: Opus 5.5 high · depthfirst |
| MiMo-V2.6-Flash | 53.5 | dfbench: MiMo-v2.6 Flash · depthfirst |
| GPT-5.6 Luna | 52.4 | dfbench: GPT 5.6 Luna xhigh · depthfirst |
| Kimi K3 | 48.0 | dfbench: Kimi K3 max · depthfirst |
| Claude Opus 5 | 47.8 | dfbench: Opus 5 med · depthfirst |
| Muse Spark 1.3 | 47.4 | dfbench: Muse Spark 1.3 xhigh · depthfirst |
| GPT-6 Luna | 45.7 | dfbench: GPT 6 Luna xhigh · depthfirst |
| GLM-5.2 | 40.7 | dfbench: GLM 5.2 xhigh · depthfirst |
| Gemini 3.6 Flash | 20.9 | dfbench: Gemini 3.6 Flash high · depthfirst |