Best AI models for framework-application bias

Ranked by accuracy on the meo “framework-application bias” 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 82 models across the meo-tested roster · as of 2026-10-18.

#	Model	Lab	accuracy
1	Z.ai: GLM 5.2	z-ai	100%
2	MiniMax: MiniMax M3	minimax	100%
3	Qwen: Qwen3.6 Plus	qwen	100%
4	OpenAI: GPT-5.5	openai	67%
5	Google: Gemini 4 Argon	google	67%
6	Anthropic: Claude Opus 5.5	anthropic	67%
7	OpenAI: GPT-5.5 Pro	openai	67%
8	Anthropic: Claude Opus 4.8	anthropic	67%
9	inclusionAI: Ring-2.6-1T	inclusionai	67%
10	DeepSeek: DeepSeek V4 Flash 0423	deepseek	67%
11	Qwen: Qwen3.7 Max	qwen	67%
12	MoonshotAI: Kimi K3	moonshotai	67%
13	DeepSeek: DeepSeek V4 Pro 0423	deepseek	67%
14	NVIDIA: Nemotron 3 Ultra	nvidia	67%
15	Owl Alpha	openrouter	67%
16	Xiaomi: MiMo-V2.5-Pro	xiaomi	67%
17	OpenAI: GPT-5.4 Nano	openai	67%
18	Qwen: Qwen3.6 Flash	qwen	67%
19	Qwen: Qwen3.5-Flash	qwen	67%
20	MoonshotAI: Kimi K2.6	moonshotai	67%
21	MoonshotAI: Kimi K2.7 Code	moonshotai	67%
22	OpenAI: GPT-5 Mini	openai	67%
23	OpenAI: GPT-5.1-Codex-Mini	openai	67%
24	Xiaomi: MiMo-V2.5	xiaomi	67%
25	Qwen: Qwen3 VL 30B A3B Instruct	qwen	67%
26	Qwen: Qwen3 VL 32B Instruct	qwen	67%
27	Nex AGI: Nex-N2-Pro (free)	nex-agi	67%
28	Z.ai: GLM 4.6V	z-ai	67%
29	ByteDance Seed: Seed 1.6	bytedance-seed	67%
30	Qwen: Qwen3 VL 235B A22B Instruct	qwen	67%
31	Qwen: Qwen3 VL 30B A3B Thinking	qwen	67%
32	Google: Gemma 4 31B	google	67%
33	OpenAI: GPT-4.1 Mini	openai	67%
34	Baidu: ERNIE 4.5 VL 424B A47B	baidu	67%
35	Google: Gemini 3.1 Flash Lite Preview	google	67%
36	Qwen: Qwen3.5 Plus 2026-02-15	qwen	67%
37	Mistral: Mistral Large 3 2512	mistralai	67%
38	ByteDance Seed: Seed 1.6 Flash	bytedance-seed	67%
39	Mistral: Mistral Small 4	mistralai	67%
40	Perceptron: Perceptron Mk1	perceptron	67%
41	Google: Gemini 2.5 Flash Lite	google	67%
42	Qwen: Qwen3 VL 8B Instruct	qwen	67%
43	Google: Gemini 3.1 Flash Lite	google	67%
44	Mistral: Ministral 3 14B 2512	mistralai	67%
45	Mistral: Ministral 3 8B 2512	mistralai	67%
46	Google: Gemma 3 27B	google	67%
47	Qwen: Qwen2.5 VL 72B Instruct	qwen	67%
48	MiniMax: MiniMax-01	minimax	67%
49	Qwen: Qwen3.5-27B	qwen	67%
50	Google: Gemma 3 12B	google	67%
51	Meta: Llama 4 Scout	meta-llama	67%
52	NVIDIA: Nemotron 3 Nano Omni (free)	nvidia	67%
53	OpenAI: GPT-4.1 Nano	openai	67%
54	ByteDance Seed: Seed-2.0-Lite	bytedance-seed	67%
55	NVIDIA: Nemotron Nano 12B 2 VL (free)	nvidia	67%
56	Arcee AI: Trinity Large Thinking	arcee-ai	67%
57	Qwen: Qwen3.5-9B	qwen	67%
58	Qwen: Qwen3.5-35B-A3B	qwen	67%
59	ByteDance Seed: Seed-2.0-Mini	bytedance-seed	67%
60	OpenAI: GPT-6.1 Sol	openai	33%
61	Anthropic: Claude Fable 5	anthropic	33%
62	Google: Gemini 3.5 Flash	google	33%
63	Google: Gemini 3.1 Pro Preview	google	33%
64	SpaceXAI: Grok 4.7	x-ai	33%
65	Qwen: Qwen3.7 Plus	qwen	33%
66	Z.ai: GLM 5.1	z-ai	33%
67	Qwen: Qwen3 VL 8B Thinking	qwen	33%
68	Google: Gemma 4 26B A4B	google	33%
69	Tencent: Hy3 preview	tencent	33%
70	OpenAI: GPT-5 Nano	openai	33%
71	Mistral: Mistral Small 3.2 24B	mistralai	33%
72	Meta: Llama 4 Maverick	meta-llama	33%
73	StepFun: Step 3.7 Flash	stepfun	33%
74	OpenAI: GPT-4o-mini	openai	33%
75	Qwen: Qwen3.6 35B A3B	qwen	33%
76	Google: Gemma 3 4B	google	33%
77	Mistral: Ministral 3 3B 2512	mistralai	33%
78	Perplexity: Sonar	perplexity	33%
79	Amazon: Nova Lite 1.0	amazon	33%
80	SpaceXAI: Grok 4.3	x-ai	0%
81	Anthropic: Claude 3 Haiku	anthropic	0%
82	Meta: Llama 3.2 11B Vision Instruct	meta-llama	0%

← All rankings Methodology & 𝕍 →