MirrorCode leaderboard: which LLMs score highest

MirrorCode (reimplementing programs from behaviour) currently has scores for 35 models. The top score is 97.8% by Claude Sonnet 5.5 (max); the median model scores 48.9% and the lowest scores 4.4%.

MirrorCode leaderboard: top 35 of 35 models (highest-effort row per model)
#ModelOrganisationMirrorCode scoreCheapest input $/MSourceCaptured
1Claude Sonnet 5.5 (max)Anthropic97.8%$2.00Epoch AI Benchmarking Hub2026-11-18
2Claude Opus 5.5 (max)Anthropic95.6%$4.00Epoch AI Benchmarking Hub2026-11-18
3Claude Fable 5.1 (max)Anthropic93.3%$10.00Epoch AI Benchmarking Hub2026-11-18
4Claude Opus 5 (max)Anthropic87.8%$5.00Epoch AI Benchmarking Hub2026-11-18
5Claude Haiku 5.5 (max)Anthropic78.9%$0.100Epoch AI Benchmarking Hub2026-11-18
6GPT-6.1 Terra (max)OpenAI76.7%$1.00Epoch AI Benchmarking Hub2026-11-18
7Claude Fable 5 (max)Anthropic70.0%$10.00Epoch AI Benchmarking Hub2026-11-18
8GPT-6.2 Astra (max)OpenAI67.8%$8.00Epoch AI Benchmarking Hub2026-11-18
9deepseek-v4.1-flash-maxDeepSeek64.4%—Epoch AI Benchmarking Hub2026-11-18
10Claude Sonnet 5 (max)Anthropic63.3%$2.00Epoch AI Benchmarking Hub2026-11-18
11GPT-6.1 Sol (max)OpenAI61.1%$2.00Epoch AI Benchmarking Hub2026-11-18
12GPT-6 Astra (max)OpenAI60.0%$10.00Epoch AI Benchmarking Hub2026-11-18
13gemini-4-argon-highGoogle58.9%$2.00Epoch AI Benchmarking Hub2026-11-18
14Grok 4.7 (xhigh)xAI54.4%$2.00Epoch AI Benchmarking Hub2026-11-18
15Kimi K3 (max)Moonshot54.4%$0.192Epoch AI Benchmarking Hub2026-11-18
16Muse Spark 1.3 (max)Meta AI54.4%$1.25Epoch AI Benchmarking Hub2026-11-18
17Qwen3.8 Max (xhigh)Alibaba52.2%$1.65Epoch AI Benchmarking Hub2026-11-18
18GPT-6 Sol (max)OpenAI48.9%$2.00Epoch AI Benchmarking Hub2026-11-18
19Qwen3.8 Max (0902) (xhigh)Alibaba47.8%$1.65Epoch AI Benchmarking Hub2026-11-18
20Muse Wonder (max)Meta AI45.6%$4.00Epoch AI Benchmarking Hub2026-11-18
21Claude Opus 4.7 (max)Anthropic43.3%$5.00Epoch AI Benchmarking Hub2026-11-18
22Gemini 3.7 Flash (high)Google DeepMind41.1%$0.750Epoch AI Benchmarking Hub2026-11-18
23GPT-5.6 Terra (max)OpenAI41.1%$2.00Epoch AI Benchmarking Hub2026-11-18
24Grok 4.6 (xhigh)xAI41.1%$2.00Epoch AI Benchmarking Hub2026-11-18
25Mistral Large 4 Preview (high)Mistral AI40.0%$0.700Epoch AI Benchmarking Hub2026-11-18
26Gemini 3.8 Flash (high)Google DeepMind38.9%$0.750Epoch AI Benchmarking Hub2026-11-18
27Claude Opus 4.6 (max)Anthropic33.3%$5.00Epoch AI Benchmarking Hub2026-11-18
28DeepSeek V4 Pro 0813 (max)DeepSeek33.3%$0.111Epoch AI Benchmarking Hub2026-11-18
29GPT-5.2 (xhigh)OpenAI32.2%$1.75Epoch AI Benchmarking Hub2026-11-18
30Grok 4.5 (high)xAI32.2%$2.00Epoch AI Benchmarking Hub2026-11-18
31GPT-6 Luna (max)OpenAI26.7%$0.100Epoch AI Benchmarking Hub2026-11-18
32GPT-5.4 (xhigh)OpenAI23.3%$2.50Epoch AI Benchmarking Hub2026-11-18
33GPT-5.5 (xhigh)OpenAI23.3%$5.00Epoch AI Benchmarking Hub2026-11-18
34GPT-5.6 Luna (max)OpenAI21.1%$0.200Epoch AI Benchmarking Hub2026-11-18
35GPT-5 (high)OpenAI4.4%$1.25Epoch AI Benchmarking Hub2026-11-18

Compare all benchmarks side by side on the live leaderboard →

Among the ten highest scorers, the cheapest listed API price belongs to Claude Haiku 5.5 (max) at $0.100 per million input tokens (score 78.9%).

What MirrorCode measures

MirrorCode measures reimplementing programs from behaviour.

How to read these scores

Scores are percentages (higher is better): the share of questions or tasks the model completed correctly. Where a model is listed at several reasoning efforts, only its highest-effort row is shown. See the methodology for how rows are chosen.

Where the data comes from

Epoch AI Benchmarking Hub (CC BY 4.0). Scores were last captured 2026-11-18; the Captured column gives each row's date.

Related benchmarks