EVMbench Incidents (Detect): leaderboard
Metric: Detect score (%): share of graded incidents (20 to 22 of the 22 real-world smart-contract incidents) whose single ground-truth vulnerability the agent's audit report identifies, graded by EVMbench's GPT-5 model-based detect grader; agents audit the deployed contracts without hints in Claude Code, Codex CLI or OpenCode (the Gemini 3.1 Pro run adds custom tools); higher is better. Source: arxiv.org. Saturation forecast: Around June 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 65 | #60 |
| 2 | GPT-5.3 Codex (High) | 59.1 | #68 (GPT-5.3 Codex) |
| 3 | Claude Sonnet 4.6 | 55 | #85 |
| 4 | GPT-5.3 Codex (xHigh) | 50 | #68 (GPT-5.3 Codex) |
| 5 | GLM-5 | 42.9 | #137 |
| 6 | Gemini 3.1 Pro (Preview) | 30 | #54 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=evmbench-incidents-detect · How It Works · Data refreshed daily, snapshot 2026-10-11.