# Vercel — DeepsecBench: evaluating model performance in finding cybersecurity vulnerabilities

- Company: Vercel (vercel.com)
- Announced: 2026-07-27T04:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/deepsecbench-evaluating-model-performance-in-finding-cybersecurity-vulnerabilities
- Record: https://forck.live/items/20683-deepsecbench-evaluating-model-performance-in-finding-cybersecurity
- Subject: v0 / AI SDK / AI Gateway
- Models affected: GPT-5.6 Sol, Claude Opus 5, Kimi K3, Grok 4.5, GPT-5.6 Luna, GPT-5.5, GPT-5.6 Terra, Gemini 3.6 Flash, GLM-5.2, Claude Opus 4.8, Inkling, Claude Haiku 4.5, Gemini 3.5 Flash Lite
- Pricing: GPT-5.6 Sol xhigh: $55.98; Claude Opus 5 medium: $31.96; Kimi K3 high: $12.38; Grok 4.5 high: $5.60; GPT-5.6 Sol medium: $17.95; GPT-5.6 Luna medium: $5.23; GPT-5.6 Terra medium: $5.15; Gemini 3.5 Flash Lite high: $0.18; costs cover 50 files

Vercel released DeepsecBench, a benchmark that evaluates how well different models find cybersecurity vulnerabilities in application code. The benchmark uses a golden set of 231 human-judged findings from an open-source codebase and reports recall, precision, cost, and total time for each model. GPT-5.6 Sol scored highest (35.58) at $55.98 on the xhigh setting, while cheaper models like Kimi K3 and Grok 4.5 offered competitive scores at lower cost.

## Evidence

Verbatim from https://vercel.com/blog/deepsecbench-evaluating-model-performance-in-finding-cybersecurity-vulnerabilities:

> Today we're releasing DeepsecBench , a benchmark that evaluates how well different models find cybersecurity vulnerabilities in application code.

---

Record: https://forck.live/items/20683-deepsecbench-evaluating-model-performance-in-finding-cybersecurity
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md