CyberChainBench - Detect: leaderboard
Smart-contract vulnerability detection on 541 real exploit incidents from DeFiHackLabs across nine EVM chains; CyberChainBench (Huang et al., 2026). An agent with on-chain tools must name the vulnerable function and the root-cause vulnerability type, and the score is the percentage of cases where both match; higher is better. Rows are agent harness and model configurations.
Metric: Detection reward (%; 1 when both the root-cause vulnerability type and the vulnerable function match the ground truth, else 0; all 541 cases; each model runs in its own agent harness (Claude Code, Codex, Gemini CLI or OpenCode) with the benchmark's on-chain MCP tools in a Harbor container, 30-minute budget, best validated attempt). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 37.5 | #48 |
| 2 | Gemini 3.1 Pro (Preview) | 37.2 | #91 |
| 3 | Claude Opus 4.7 | 33.8 | #75 |
| 4 | GPT-5.4 | 32.5 | #121 |
| 5 | Claude Opus 4.6 | 30.9 | #100 |
| 6 | DeepSeek V4 Pro | 22.6 | #181 |
| 7 | GPT-5.2 | 17.7 | #172 |
| 8 | GLM-5.1 | 16.8 | #175 |
| 9 | Kimi K2.6 | 6.5 | #173 |
| 10 | MiniMax-M2.7 | 4.6 | #519 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cyberchainbench-detect · How It Works · Data refreshed daily, snapshot 2026-11-12.