Epoch AI - Mirrorcode: leaderboard
MirrorCode (Epoch AI) — long-horizon autonomous coding: agents reimplement entire programs (~25 targets across Unix utilities, interpreters, cryptography, bioinformatics) from documentation and a runnable executable only, with no access to the original source. Scored as the fraction of the target's behavior reproduced.
Metric: Score. Source: epoch.ai. Saturation forecast: Around May 2028. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 (Max) | 77.4 | #3 (Claude Opus 5.5) |
| 2 | Claude Fable 5.1 (High) | 73.3 | #9 (Claude Fable 5.1) |
| 3 | Claude Fable 5 (High) | 63.89 | #16 (Claude Fable 5) |
| 4 | GPT-6 Astra (High) | 46.7 | #6 (GPT-6 Astra) |
| 5 | Claude Opus 4.7 (High) | 31.11 | #49 (Claude Opus 4.7) |
| 6 | GPT-5.6 Sol (High) | 20 | #19 (GPT-5.6 Sol) |
| 7 | GPT-5.4 (2026-03-05) (High) | 15.56 | #59 (GPT-5.4 (2026-03-05)) |
| 8 | GPT-5.5 (High) | 10 | #31 (GPT-5.5) |
| 9 | Gemini 3.1 Pro (Preview) (High) | 8.89 | #58 (Gemini 3.1 Pro (Preview)) |
No result here: #7 GPT-6.1 Sol, #8 Gemini 4 Argon, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=epoch-ai-mirrorcode · How It Works · Data refreshed daily, snapshot 2026-10-20.