Best AI models for perceptual illusions

Ranked by accuracy on the meo “perceptual illusions” domain — one of eleven un-leaked, contamination-resistant domains. Per-domain strength reveals where a model genuinely reasons versus where it pattern-matches; the leaders differ sharply by domain.

Ranking 71 models across the meo-tested roster · as of 2026-10-18.

#	Model	Lab	accuracy
1	OpenAI: GPT-5.5	openai	100%
2	Anthropic: Claude Opus 5.5	anthropic	100%
3	OpenAI: GPT-6.1 Sol	openai	100%
4	Google: Gemini 4 Argon	google	99%
5	OpenAI: GPT-5.5 Pro	openai	98%
6	Anthropic: Claude Fable 5	anthropic	89%
7	Google: Gemini 3.5 Flash	google	80%
8	Anthropic: Claude Opus 4.8	anthropic	78%
9	Google: Gemini 3.1 Pro Preview	google	75%
10	MoonshotAI: Kimi K3	moonshotai	70%
11	Z.ai: GLM 4.6V	z-ai	63%
12	Meta: Llama 4 Maverick	meta-llama	63%
13	SpaceXAI: Grok 4.7	x-ai	59%
14	OpenAI: GPT-4.1 Nano	openai	59%
15	Baidu: ERNIE 4.5 VL 424B A47B	baidu	58%
16	Google: Gemma 3 12B	google	58%
17	Google: Gemma 3 4B	google	57%
18	Perceptron: Perceptron Mk1	perceptron	56%
19	Google: Gemini 2.5 Flash Lite	google	56%
20	StepFun: Step 3.7 Flash	stepfun	56%
21	OpenAI: GPT-4o-mini	openai	56%
22	NVIDIA: Nemotron Nano 12B 2 VL (free)	nvidia	56%
23	Qwen: Qwen3.5-9B	qwen	56%
24	Qwen: Qwen3.7 Plus	qwen	55%
25	OpenAI: GPT-4.1 Mini	openai	55%
26	Meta: Llama 4 Scout	meta-llama	55%
27	MiniMax: MiniMax M3	minimax	54%
28	SpaceXAI: Grok 4.3	x-ai	54%
29	Qwen: Qwen3 VL 30B A3B Instruct	qwen	54%
30	Google: Gemini 3.1 Flash Lite Preview	google	54%
31	MiniMax: MiniMax-01	minimax	54%
32	ByteDance Seed: Seed-2.0-Lite	bytedance-seed	54%
33	Qwen: Qwen3.6 35B A3B	qwen	54%
34	Qwen: Qwen3.6 Flash	qwen	53%
35	Qwen: Qwen3 VL 32B Instruct	qwen	53%
36	ByteDance Seed: Seed 1.6	bytedance-seed	53%
37	Google: Gemini 3.1 Flash Lite	google	53%
38	Anthropic: Claude 3 Haiku	anthropic	53%
39	Qwen: Qwen3.5-Flash	qwen	52%
40	MoonshotAI: Kimi K2.7 Code	moonshotai	52%
41	OpenAI: GPT-5 Mini	openai	52%
42	OpenAI: GPT-5.1-Codex-Mini	openai	52%
43	Xiaomi: MiMo-V2.5	xiaomi	52%
44	Qwen: Qwen3 VL 30B A3B Thinking	qwen	52%
45	Qwen: Qwen3.6 Plus	qwen	52%
46	MoonshotAI: Kimi K2.6	moonshotai	51%
47	Nex AGI: Nex-N2-Pro (free)	nex-agi	51%
48	Qwen: Qwen3 VL 235B A22B Instruct	qwen	51%
49	Qwen: Qwen3.5 Plus 2026-02-15	qwen	51%
50	Mistral: Mistral Large 3 2512	mistralai	51%
51	Mistral: Mistral Small 4	mistralai	51%
52	Qwen: Qwen2.5 VL 72B Instruct	qwen	51%
53	Qwen: Qwen3.5-27B	qwen	51%
54	ByteDance Seed: Seed-2.0-Mini	bytedance-seed	51%
55	ByteDance Seed: Seed 1.6 Flash	bytedance-seed	50%
56	Qwen: Qwen3 VL 8B Instruct	qwen	50%
57	Mistral: Ministral 3 8B 2512	mistralai	50%
58	Perplexity: Sonar	perplexity	50%
59	Mistral: Mistral Small 3.2 24B	mistralai	49%
60	NVIDIA: Nemotron 3 Nano Omni (free)	nvidia	49%
61	OpenAI: GPT-5.4 Nano	openai	48%
62	OpenAI: GPT-5 Nano	openai	48%
63	Google: Gemma 3 27B	google	48%
64	Mistral: Ministral 3 3B 2512	mistralai	48%
65	Amazon: Nova Lite 1.0	amazon	48%
66	Qwen: Qwen3.5-35B-A3B	qwen	48%
67	Google: Gemma 4 26B A4B	google	47%
68	Meta: Llama 3.2 11B Vision Instruct	meta-llama	47%
69	Google: Gemma 4 31B	google	46%
70	Qwen: Qwen3 VL 8B Thinking	qwen	46%
71	Mistral: Ministral 3 14B 2512	mistralai	45%

← All rankings Methodology & 𝕍 →