Vals AI ReverseEngBench: leaderboard
Vals AI agentic reverse-engineering benchmark (formerly SRE Bench): agents are given real-world, contamination-free binaries and must recover their behaviour; the headline score is the share of binaries fully solved, alongside a partial-credit capability score.
Metric: Fully Solved (%). Source: www.vals.ai. Saturation forecast: Around January 2027. 35 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 56.87 | #6 |
| 2 | GPT-6.1 Sol | 50.76 | #7 |
| 3 | Gemini 4 Argon | 44.27 | #8 |
| 4 | Claude Opus 5.5 | 33.59 | #3 |
| 5 | GPT-5.6 Sol | 30.53 | #19 |
| 6 | GPT-6 Sol | 30.53 | #14 |
| 7 | Claude Sonnet 5.5 | 30.15 | #11 |
| 8 | Claude Fable 5.1 | 22.9 | #9 |
| 9 | Claude Opus 5 | 12.21 | #12 |
| 10 | Muse Wonder Full | 11.83 | #10 |
| 11 | Gemini 3.8 Flash | 11.07 | #21 |
| 12 | GPT-5.6 Terra | 9.54 | #41 |
| 13 | GLM-5.3 | 9.16 | #39 |
| 14 | Claude Haiku 5.5 | 5.72 | #40 |
| 15 | Gemini 3.7 Flash | 4.58 | #23 |
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-reverseengbench · How It Works · Data refreshed daily, snapshot 2026-10-20.