dfbench Differential Analysis
Cybersecurity · 2026-08-19
Macro recall across new, kept and closed vulnerabilities as dfbench moves a scope from a base commit to a target commit, given the known vulnerabilities at the base.
Benchmark source: Depthfirst
Top models (higher is better)
| Model | Score | Source |
|---|---|---|
| Claude Opus 5.5 | 79.5 | dfbench: Opus 5.5 high · depthfirst |
| MiMo-V2.6-Pro | 76.5 | dfbench: MiMo-v2.6 Pro · depthfirst |
| GPT-5.6 Sol | 76.3 | dfbench: GPT 5.6 Sol high · depthfirst |
| GPT-6 Sol | 76.3 | dfbench: GPT 6 Sol high · depthfirst |
| Grok 4.7 | 74.8 | dfbench: Grok 4.7 xhigh · depthfirst |
| Muse Spark 1.3 | 72.8 | dfbench: Muse Spark 1.3 xhigh · depthfirst |
| MiMo-V2.6-Flash | 72.7 | dfbench: MiMo-v2.6 Flash · depthfirst |
| GPT-6 Luna | 72.4 | dfbench: GPT 6 Luna high · depthfirst |
| Claude Opus 5 | 71.3 | dfbench: Opus 5 med · depthfirst |
| GPT-5.6 Luna | 71.3 | dfbench: GPT 5.6 Luna high · depthfirst |
| DeepSeek V4.1 Flash | 70.1 | dfbench: DeepSeek v4.1 Flash high · depthfirst |
| Grok 4.5 | 65.1 | dfbench: Grok 4.5 high · depthfirst |
| GLM-5.2 | 61.4 | dfbench: GLM 5.2 xhigh · depthfirst |
| Gemini 3.6 Flash | 61.0 | dfbench: Gemini 3.6 Flash high · depthfirst |
| Kimi K3 | 60.3 | dfbench: Kimi K3 max · depthfirst |