benchgap
General

CompileBench leaderboard

As of 2026-11-18, the highest measured score on CompileBench is 86.7% by Claude Sonnet 4.5. 51 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Haiku 5.5100.0%estimated ± 10.0 pp, low confidence
2Claude Opus 4.7100.0%estimated ± 10.0 pp, low confidence
3Claude Opus 4.8100.0%estimated ± 10.0 pp, low confidence
4Claude Sonnet 5100.0%estimated ± 10.0 pp, low confidence
5DeepSeek V4 Flash 0731100.0%estimated ± 10.0 pp, low confidence
6Gemini 3.1 Pro100.0%estimated ± 10.0 pp, low confidence
7Gemini 3.5 Flash100.0%estimated ± 10.0 pp, low confidence
8Gemini 3.6 Flash100.0%estimated ± 10.0 pp, low confidence
9Gemini 3.7 Flash100.0%estimated ± 10.0 pp, low confidence
10GLM-5.3100.0%estimated ± 10.0 pp, low confidence
11GPT-5.4100.0%estimated ± 10.0 pp, low confidence
12GPT-5.6 Terra100.0%estimated ± 10.0 pp, low confidence
13Grok 4.6100.0%estimated ± 10.0 pp, low confidence
14Grok 4.7100.0%estimated ± 10.0 pp, low confidence
15Kimi K3.1 Pro100.0%estimated ± 10.0 pp, low confidence
16o3100.0%estimated ± 10.0 pp, low confidence
17Qwen3.7 Max100.0%estimated ± 10.0 pp, low confidence
18Qwen3.8 Max100.0%estimated ± 10.0 pp, low confidence
19Kimi K397.7%estimated ± 10.0 pp, low confidence
20Claude Opus 4.695.5%estimated ± 10.0 pp, low confidence
21Muse Spark 1.395.5%estimated ± 10.0 pp, low confidence
22GPT-5.291.1%estimated ± 10.0 pp, low confidence
23Claude Opus 4.588.9%estimated ± 10.0 pp, low confidence
24Qwen3.6-35B-A3B88.9%estimated ± 10.0 pp, low confidence
25Claude Sonnet 4.586.7%measured
26GPT-586.7%measured
27GPT-5.6 Luna86.6%estimated ± 10.0 pp, low confidence
28Gemini 3 Flash84.4%estimated ± 10.0 pp, low confidence
29Nemotron 3 Ultra84.4%estimated ± 10.0 pp, low confidence
30Qwen3.5 Flash84.4%estimated ± 10.0 pp, low confidence
31Claude Sonnet 4.5 Thinking84.4%measured
32Gemini 3.5 Flash-Lite82.2%estimated ± 10.0 pp, low confidence
33Qwen 3.6 Max (preview)82.2%estimated ± 10.0 pp, low confidence
34Kimi K2.680.0%estimated ± 10.0 pp, low confidence
35Qwen3.5 397B80.0%estimated ± 10.0 pp, low confidence
36Mistral Large 477.8%estimated ± 10.0 pp, low confidence
37Qwen3.5-122B-A10B77.8%estimated ± 10.0 pp, low confidence
38Qwen3.5 Plus77.8%estimated ± 10.0 pp, low confidence
39Qwen3.7 Plus77.8%estimated ± 10.0 pp, low confidence
40Claude Sonnet 4.675.5%estimated ± 10.0 pp, low confidence
41GPT-5.175.5%estimated ± 10.0 pp, low confidence
42GLM-5.273.3%estimated ± 10.0 pp, low confidence
43Qwen3.7 Flash73.3%estimated ± 10.0 pp, low confidence
44Claude Haiku 4.5 Thinking73.3%measured
45Grok 471.1%measured
46Claude Haiku 4.568.9%measured
47GPT-4o mini66.7%estimated ± 10.0 pp, low confidence
48Qwen3.6 Plus66.7%estimated ± 10.0 pp, low confidence
49DeepSeek V3.164.4%measured
50Grok Code Fast 164.4%measured
51GLM-5.3-Flash57.8%estimated ± 10.0 pp, low confidence
52MiniMax M357.8%estimated ± 10.0 pp, low confidence
53Grok 4 Fast (Reasoning)55.6%measured
54GPT-5.4 mini55.6%estimated ± 10.0 pp, low confidence
55GPT-6 Luna55.6%estimated ± 10.0 pp, low confidence
56o3-mini55.6%estimated ± 10.0 pp, low confidence
57Qwen3.6-27B55.6%estimated ± 10.0 pp, low confidence
58Inkling-Small53.4%estimated ± 10.0 pp, low confidence
59Gemini 2.5 Pro53.3%measured
60GPT-4.153.3%measured
61Qwen3 Max53.3%measured
62GPT-5.4 nano51.1%estimated ± 10.0 pp, low confidence
63o4-mini51.1%estimated ± 10.0 pp, low confidence
64GLM-4.548.9%measured
65GPT-4.1 mini46.7%measured
66GPT-OSS 120B42.2%measured
67Gemini 2.5 Flash40.0%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General