CyberChainBench - Exploit: leaderboard
Exploit generation on CyberChainBench (Huang et al., 2026): the agent writes a Foundry exploit for a real DeFi incident, executed on a fork of the chain at the historical block. The score is the mean of realised profit divided by the historical attacker profit, capped at 1, in percent over the 200-case exploit set; higher is better.
Metric: Exploit reward (%; min(agent profit / reference profit, 1) for a Foundry exploit executed on a historical mainnet fork, 200-case exploit set; 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 | 43.7 | #48 |
| 2 | GPT-5.4 | 43.3 | #121 |
| 3 | Claude Opus 4.7 | 42.3 | #75 |
| 4 | Claude Opus 4.6 | 37.4 | #100 |
| 5 | DeepSeek V4 Pro | 36.6 | #181 |
| 6 | GLM-5.1 | 24.9 | #175 |
| 7 | Gemini 3.1 Pro (Preview) | 20.2 | #91 |
| 8 | GPT-5.2 | 17.5 | #172 |
| 9 | MiniMax-M2.7 | 12 | #519 |
| 10 | Kimi K2.6 | 3.6 | #173 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cyberchainbench-exploit · How It Works · Data refreshed daily, snapshot 2026-11-12.