benchgap
General

Cookie-Bench leaderboard

As of 2026-11-18, the highest measured score on Cookie-Bench is 91.5% by Claude Opus 5.5. 98 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Opus 5.591.5%measured
2GPT-6.2 Astra90.8%estimated ± 1.3 pp, low confidence
3Claude Fable 5.290.8%estimated ± 1.3 pp, low confidence
4GPT-6 Astra90.7%estimated ± 1.3 pp, low confidence
5GPT-6.1 Terra90.6%estimated ± 1.3 pp, medium confidence
6Gemini 4 Carbon90.5%estimated ± 1.3 pp, medium confidence
7Gemini 3 Pro90.3%estimated ± 4.8 pp, medium confidence
8Claude Sonnet 5.590.2%estimated ± 1.3 pp, medium confidence
9Gemini 4 Argon90.2%estimated ± 1.3 pp, medium confidence
10Claude Opus 4.6 (Adaptive)89.8%estimated ± 4.8 pp, medium confidence
11Claude Opus 589.7%estimated ± 1.3 pp, medium confidence
12GPT-6.1 Sol89.7%measured
13Claude Fable 5.189.6%estimated ± 1.3 pp, medium confidence
14GPT-5.6 Sol89.6%estimated ± 1.3 pp, medium confidence
15GPT-5.5 Pro89.5%estimated ± 2.5 pp, medium confidence
16GPT-5.589.4%estimated ± 1.3 pp, medium confidence
17GPT-6 Sol89.4%estimated ± 1.3 pp, medium confidence
18Grok 4.20 Multi-agent89.2%estimated ± 4.8 pp, medium confidence
19Hy488.8%measured
20Claude Fable 588.7%estimated ± 1.3 pp, medium confidence
21GPT-5.4 Pro88.6%estimated ± 2.5 pp, medium confidence
22Grok 4.1 Fast (Reasoning)88.5%estimated ± 4.8 pp, medium confidence
23Muse Wonder88.4%estimated ± 1.3 pp, medium confidence
24ERNIE 5.5 Flash Preview88.2%measured
25DeepSeek V4.1 Flash 101787.9%measured
26Gemini 3.8 Flash87.3%estimated ± 1.3 pp, medium confidence
27Kimi K3.1 Pro85.2%estimated ± 1.3 pp, medium confidence
28Muse Spark Open84.5%estimated ± 4.8 pp, medium confidence
29Step 5 Preview83.3%estimated ± 4.8 pp, medium confidence
30Qwen3.8 Max83.1%estimated ± 1.3 pp, medium confidence
31Gemini 3.7 Flash82.6%estimated ± 1.3 pp, medium confidence
32GPT-5.482.6%estimated ± 1.3 pp, medium confidence
33Claude Opus 4.882.0%estimated ± 1.3 pp, medium confidence
34Claude Sonnet 581.5%estimated ± 1.3 pp, medium confidence
35GPT-5.6 Terra81.5%estimated ± 1.3 pp, medium confidence
36Grok 4.581.5%estimated ± 2.5 pp, medium confidence
37Muse Spark 1.181.2%estimated ± 2.5 pp, medium confidence
38DeepSeek V4 Flash 073181.0%estimated ± 1.3 pp, medium confidence
39Gemini 3.1 Pro81.0%estimated ± 1.3 pp, medium confidence
40Grok 4.681.0%estimated ± 1.3 pp, medium confidence
41GLM-5.380.4%measured
42Gemini 3.5 Flash80.0%estimated ± 1.3 pp, medium confidence
43Qwen3.7 Max80.0%estimated ± 1.3 pp, medium confidence
44Grok 4.2079.7%estimated ± 2.5 pp, medium confidence
45Claude Haiku 5.579.2%estimated ± 1.3 pp, medium confidence
46Gemini 3.6 Flash79.2%estimated ± 1.3 pp, medium confidence
47Muse Spark 1.279.1%estimated ± 2.5 pp, medium confidence
48Grok 4.778.8%estimated ± 1.3 pp, medium confidence
49o378.8%estimated ± 1.3 pp, medium confidence
50Seed 2.1 Turbo78.8%estimated ± 2.5 pp, medium confidence
51Claude Opus 4.778.5%estimated ± 1.3 pp, medium confidence
52Gemma 4 31B78.0%estimated ± 4.8 pp, low confidence
53Kimi K377.9%estimated ± 1.3 pp, medium confidence
54Claude Opus 4.677.7%estimated ± 1.3 pp, medium confidence
55Muse Spark 1.377.7%estimated ± 1.3 pp, medium confidence
56GPT-577.4%estimated ± 1.3 pp, medium confidence
57GPT-5.277.4%estimated ± 1.3 pp, medium confidence
58GLM-4.777.3%estimated ± 2.5 pp, medium confidence
59Inkling77.3%estimated ± 2.5 pp, medium confidence
60Seed 2.1 Pro77.3%estimated ± 2.5 pp, medium confidence
61Claude Opus 4.577.3%estimated ± 1.3 pp, medium confidence
62Qwen3.6-35B-A3B77.3%estimated ± 1.3 pp, medium confidence
63GPT-5.6 Luna77.2%estimated ± 1.3 pp, medium confidence
64Gemini 3 Flash77.1%estimated ± 1.3 pp, medium confidence
65Nemotron 3 Ultra77.1%estimated ± 1.3 pp, medium confidence
66Qwen3.5 Flash77.1%estimated ± 1.3 pp, medium confidence
67Gemini 3.5 Flash-Lite77.0%estimated ± 1.3 pp, medium confidence
68Qwen 3.6 Max (preview)77.0%estimated ± 1.3 pp, medium confidence
69GLM-5.177.0%estimated ± 2.5 pp, low confidence
70Kimi K2.677.0%estimated ± 1.3 pp, medium confidence
71Qwen3.5 397B77.0%estimated ± 1.3 pp, medium confidence
72Claude Sonnet 4.577.0%estimated ± 1.3 pp, medium confidence
73Mistral Large 477.0%estimated ± 1.3 pp, medium confidence
74Qwen3.5-122B-A10B77.0%estimated ± 1.3 pp, medium confidence
75Qwen3.5 Plus77.0%estimated ± 1.3 pp, medium confidence
76Qwen3.7 Plus77.0%estimated ± 1.3 pp, medium confidence
77Claude Sonnet 4.676.9%estimated ± 1.3 pp, medium confidence
78GPT-5.176.9%estimated ± 1.3 pp, medium confidence
79GLM-5.276.9%estimated ± 1.3 pp, medium confidence
80Qwen3.7 Flash76.9%estimated ± 1.3 pp, medium confidence
81GLM-5.3-Flash76.9%measured
82GPT-4o mini76.9%estimated ± 1.3 pp, medium confidence
83Qwen3.6 Plus76.9%estimated ± 1.3 pp, medium confidence
84MiniMax M376.9%estimated ± 1.3 pp, medium confidence
85GPT-4.1 mini76.9%estimated ± 1.3 pp, low confidence
86GPT-5.4 mini76.9%estimated ± 1.3 pp, low confidence
87GPT-6 Luna76.9%estimated ± 1.3 pp, low confidence
88o3-mini76.9%estimated ± 1.3 pp, low confidence
89Qwen3.6-27B76.9%estimated ± 1.3 pp, low confidence
90Inkling-Small76.9%estimated ± 1.3 pp, low confidence
91GPT-5.4 nano76.9%estimated ± 1.3 pp, low confidence
92o4-mini76.9%estimated ± 1.3 pp, low confidence
93Qwen3 Max76.9%estimated ± 1.3 pp, low confidence
94GPT-OSS 120B76.9%estimated ± 1.3 pp, low confidence
95Grok 4.376.4%estimated ± 2.5 pp, low confidence
96Kimi K2.7 Code75.2%estimated ± 2.5 pp, low confidence
97Hy4 preview75.2%estimated ± 4.8 pp, low confidence
98GPT-4.174.3%estimated ± 2.5 pp, low confidence
99o174.0%estimated ± 2.5 pp, low confidence
100GPT-4o73.7%estimated ± 2.5 pp, low confidence
101Claude Haiku 4.573.4%estimated ± 2.5 pp, low confidence
102Claude 3 Opus72.2%estimated ± 2.5 pp, low confidence
103GPT-4.1 nano71.0%estimated ± 2.5 pp, low confidence
104Qwen3.8-27B48.4%estimated ± 4.8 pp, low confidence
105Claude Opus 4.5 Thinking43.2%estimated ± 4.8 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General