Benchmark · Coding
MirrorCode
An agent rebuilds a whole program from its behaviour alone, without the source or the internet, and must match the original's output on held-out end-to-end tests. Epoch AI and METR built it; the leaderboard runs 15 medium and large programs in two languages, three attempts each, with up to 10 billion tokens and 7 days an attempt.
- Score
- % of tasks solved, higher is better
- Run by
- Epoch AI
- Status
- Current, newest result 17 November 2026
- Updated
- as models are released
- Ceiling
- The best result is within five points of a perfect score, so the test no longer tells the top models apart
- Data
- the published file, CC BY 4.0, read 20 November 2026
- Method
- epoch.ai/MirrorCode
Results
19 models we list, each at its best setting, best first.
| # | Model | Score | Range | Who ran it | Setting | Date | |
|---|---|---|---|---|---|---|---|
| 1 | Claude Sonnet 5.5 A | 97.8% | Independent | effort max | 2026-10-05 | source | |
| 2 | Claude Opus 5.5 A | 95.6% | Independent | effort max | 2026-10-30 | source | |
| 3 | Claude Fable 5.1 A | 93.3% | Independent | effort max | 2026-09-09 | source | |
| 4 | Claude Haiku 5.5 A | 78.9% | Independent | effort max | 2026-10-21 | source | |
| 5 | GPT-6.1 Terra AA | 76.7% | Independent | effort max | 2026-10-30 | source | |
| 6 | GPT-6.2 Astra AA | 67.8% | Independent | effort max | 2026-10-27 | source | |
| 7 | DeepSeek V4.1 Flash C | 64.4% | Independent | effort max | 2026-10-21 | source | |
| 8 | GPT-6.1 Sol AA | 61.1% | Independent | effort max | 2026-10-05 | source | |
| 9 | GPT-6 Astra A | 60.0% | Independent | effort max | 2026-10-29 | source | |
| 10 | Gemini 4 Argon B | 58.9% | Independent | effort high | 2026-10-14 | source | |
| 11 | Grok 4.7 BB | 54.4% | Independent | effort xhigh | 2026-10-29 | source | |
| 11 | Muse Spark 1.3 D | 54.4% | Independent | effort max | 2026-10-20 | source | |
| 11 | Kimi K3 B | 54.4% | Independent | effort max | 2026-09-03 | source | |
| 14 | GPT-6 Sol A | 48.9% | Independent | effort max | 2026-10-29 | source | |
| 15 | Muse Wonder D | 45.6% | Independent | effort max | 2026-10-14 | source | |
| 16 | Kimi K3.1 Pro D | 43.3% | Independent | effort max | 2026-11-17 | source | |
| 17 | Gemini 3.8 Flash AA | 38.9% | Independent | effort high | 2026-09-02 | source | |
| 18 | GPT-6 Luna A | 26.7% | Independent | effort max | 2026-10-21 | source | |
| 19 | Gemini 3.1 Pro (preview) AA | 8.9% | Independent | effort high | 2026-08-10 | source |
Every setting, 22 rows
| Model | As the leaderboard names it | Score | Who ran it | Setting | Date | |
|---|---|---|---|---|---|---|
| Claude Sonnet 5.5 | claude-sonnet-5-5 | 97.8% | Independent | effort max | 2026-10-05 | source |
| Claude Opus 5.5 | claude-opus-5-5 | 95.6% | Independent | effort max | 2026-10-30 | source |
| Claude Fable 5.1 | claude-fable-5-1 | 93.3% | Independent | effort max | 2026-09-09 | source |
| Claude Haiku 5.5 | claude-haiku-5-5 | 78.9% | Independent | effort max | 2026-10-21 | source |
| Claude Opus 5.5 | claude-opus-5-5 | 78.9% | Independent | effort high | 2026-09-22 | source |
| GPT-6.1 Terra | gpt-6.1-terra | 76.7% | Independent | effort max | 2026-10-30 | source |
| Claude Fable 5.1 | claude-fable-5-1 | 73.3% | Independent | effort high | 2026-09-10 | source |
| GPT-6.2 Astra | gpt-6.2-astra | 67.8% | Independent | effort max | 2026-10-27 | source |
| DeepSeek V4.1 Flash | deepseek-v4.1-flash | 64.4% | Independent | effort max | 2026-10-21 | source |
| GPT-6.1 Sol | gpt-6.1-sol | 61.1% | Independent | effort max | 2026-10-05 | source |
| GPT-6 Astra | gpt-6-astra | 60.0% | Independent | effort max | 2026-10-29 | source |
| Gemini 4 Argon | gemini-4-argon | 58.9% | Independent | effort high | 2026-10-14 | source |
| Grok 4.7 | grok-4.7 | 54.4% | Independent | effort xhigh | 2026-10-29 | source |
| Muse Spark 1.3 | muse-spark-1.3 | 54.4% | Independent | effort max | 2026-10-20 | source |
| Kimi K3 | kimi-k3 | 54.4% | Independent | effort max | 2026-09-03 | source |
| GPT-6 Sol | gpt-6-sol | 48.9% | Independent | effort max | 2026-10-29 | source |
| GPT-6 Astra | gpt-6-astra | 46.7% | Independent | effort high | 2026-08-30 | source |
| Muse Wonder | muse-wonder-1.0 | 45.6% | Independent | effort max | 2026-10-14 | source |
| Kimi K3.1 Pro | kimi-k3.1-pro-together | 43.3% | Independent | effort max | 2026-11-17 | source |
| Gemini 3.8 Flash | gemini-3.8-flash | 38.9% | Independent | effort high | 2026-09-02 | source |
| GPT-6 Luna | gpt-6-luna | 26.7% | Independent | effort max | 2026-10-21 | source |
| Gemini 3.1 Pro (preview) | gemini-3.1-pro-preview | 8.9% | Independent | effort high | 2026-08-10 | source |
A rank is among the models we list with a result, not a place on the whole leaderboard, and a model offered by several hosts is shown once, on its own listing. These results don't move our grades. Data: Epoch AI, Capabilities and benchmarking (CC BY 4.0), CC BY 4.0.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.