Skip to content

Benchmark

SnakeBench

Scores are percentages.

Models measured
260
Scores recorded
267
Sources
1
IQ weight
1.32

Score distribution

22.1%
Mean
22.5%
Median
7.3%
Std dev
13.2%
10th percentile
32.2%
90th percentile
1.2%
Min
1.2%38.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.32 for chat models, measured over 254 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1GPT 5.5· SnakeBench data38.5%
2GPT-5.1 Codex Max· SnakeBench data38.2%
3Gemini 3.1 Pro Preview· SnakeBench data38.2%
4Gemini 3 Pro Preview· SnakeBench data38%
5GPT-6.2 Astra· SnakeBench data37.8%
6GPT-6 Astra· SnakeBench data37.1%
7GPT-5.1 Codex Mini· SnakeBench data36%
8OpenAI o3· SnakeBench data35.5%
9Claude Opus 4.8· SnakeBench data35.4%
10Gemini Pro Latest· SnakeBench data34.8%
11GPT 5· SnakeBench data34.7%
12o4 Mini· SnakeBench data34.5%
13Qwen3.7 Max· SnakeBench data34.1%
14GPT-5.3 Codex· SnakeBench data33.4%
15GPT-5 Codex· SnakeBench data33.3%
16Grok 4 Fast· SnakeBench data33.2%
17Gemini 2.0 Flash· SnakeBench data33.2%
18Claude Opus 4.6· SnakeBench data33.2%
19Gemini 3.5 Flash· SnakeBench data33.1%
20Qwen3.5 122B-A10B· SnakeBench data33%
21Claude Sonnet 4.5· SnakeBench data33%
22GPT 5.1· SnakeBench data33%
23qwen3-vl-30bthinking· SnakeBench data32.5%
24GPT-5.2 Codex· SnakeBench data32.4%
25Qwen3.5 Flash· SnakeBench data32.4%
26GPT 5.2· SnakeBench data32.3%
27Claude Sonnet 4· SnakeBench data32.2%
28GPT-5.1 Codex· SnakeBench data32.2%
29Gemini 3 Flash· SnakeBench data31.9%
30Mercury· SnakeBench data31.9%
31Gemini 2.5 Flash· SnakeBench data31.3%
32GPT OSS Safeguard 20B· SnakeBench data31.2%
33GPT 4.1· SnakeBench data31.2%
34Qwen3.5 27B· SnakeBench data30.9%
35Qwen3.6 Max Preview· SnakeBench data30.9%
36Qwen3 Max· SnakeBench data30.8%
37gemini-2.5-flash-preview-09-2025· SnakeBench data30.8%
38Gemini 3.1 Flash Lite· SnakeBench data30.7%
39GLM 5· SnakeBench data30.5%
40GPT-5 Image· SnakeBench data30.5%
41Gemini 2.5 Flash Image· SnakeBench data30.1%
42Claude Opus 4.5· SnakeBench data30.1%
43Claude Opus 4.7· SnakeBench data30.1%
44Kimi K2 0905· SnakeBench data30.1%
45GPT-5.6 Terra· SnakeBench data29.8%
46o3 Deep Research· SnakeBench data29.7%
47Qwen3 235B A22B Instruct 2507· SnakeBench data29.4%
48GPT 5.4· SnakeBench data28.8%
49Gemini 2.5 Pro Preview 05-06· SnakeBench data28.8%
50Kimi K2 Thinking· SnakeBench data28.8%

Showing 1–50 of 267 results.

Data credits: SnakeBench (Greg Kamradt / ARC Prize Foundation), MIT (licence)