benchgap
General

MirrorCode leaderboard

As of 2026-11-18, the highest measured score on MirrorCode is 97.8% by Claude Sonnet 5.5. 5 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Sonnet 5.597.8%measured
2Claude Opus 5.595.6%measured
3Claude Fable 5.193.3%measured
4Claude Opus 587.8%measured
5Claude Haiku 5.578.9%measured
6GPT-6.1 Terra76.7%measured
7Gemini 4 Carbon70.4%estimated ± 15.0 pp, low confidence
8Claude Fable 570.0%measured
9GPT-6.2 Sol68.2%estimated ± 14.2 pp, low confidence
10GPT-6 Cyber68.2%estimated ± 14.2 pp, low confidence
11GPT-6.2 Astra67.8%measured
12Claude Opus 4.865.6%measured
13DeepSeek V4.1 Flash 101764.4%measured
14Claude Sonnet 563.3%measured
15GPT-6.1 Sol61.1%measured
16GPT-6 Astra60.0%measured
17Gemini 4 Argon58.9%measured
18Grok 4.754.4%measured
19Kimi K354.4%measured
20Muse Spark 1.354.4%measured
21Qwen3.8 Max52.2%measured
22GPT-6 Sol48.9%measured
23Muse Wonder45.6%measured
24MiMo-V2.6-Pro44.3%estimated ± 15.0 pp, low confidence
25Claude Opus 4.743.3%measured
26Gemini 3.7 Flash41.1%measured
27GPT-5.6 Terra41.1%measured
28Grok 4.641.1%measured
29Mistral Large 440.0%measured
30Gemini 3.8 Flash38.9%measured
31GLM-5.236.7%measured
32Claude Opus 4.633.3%measured
33DeepSeek V4 Pro33.3%measured
34GPT-5.232.2%measured
35Grok 4.532.2%measured
36MiMo-V2.6-Flash27.8%estimated ± 15.0 pp, low confidence
37GPT-6 Luna26.7%measured
38Gemini 3.6 Flash23.3%measured
39GPT-5.423.3%measured
40GPT-5.523.3%measured
41GPT-5.6 Luna21.1%measured
42GPT-5.6 Sol20.0%measured
43Gemini 3.1 Pro8.9%measured
44GPT-54.4%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General