Best AI models for base-rate bias

Ranked by accuracy on the meo “base-rate 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	Owl Alpha	openrouter	93%
2	inclusionAI: Ring-2.6-1T	inclusionai	88%
3	Qwen: Qwen3.6 Flash	qwen	88%
4	Google: Gemini 3.1 Flash Lite	google	88%
5	Tencent: Hy3 preview	tencent	88%
6	Anthropic: Claude Opus 4.8	anthropic	85%
7	DeepSeek: DeepSeek V4 Pro 0423	deepseek	85%
8	MiniMax: MiniMax M3	minimax	85%
9	Qwen: Qwen3.7 Plus	qwen	85%
10	Nex AGI: Nex-N2-Pro (free)	nex-agi	85%
11	Z.ai: GLM 5.1	z-ai	85%
12	Perceptron: Perceptron Mk1	perceptron	85%
13	Google: Gemini 3.5 Flash	google	83%
14	DeepSeek: DeepSeek V4 Flash 0423	deepseek	83%
15	Qwen: Qwen3.7 Max	qwen	83%
16	Google: Gemini 3.1 Pro Preview	google	83%
17	Xiaomi: MiMo-V2.5	xiaomi	83%
18	Qwen: Qwen3 VL 30B A3B Instruct	qwen	83%
19	Mistral: Ministral 3 14B 2512	mistralai	83%
20	Qwen: Qwen3.6 Plus	qwen	83%
21	OpenAI: GPT-5.5	openai	80%
22	OpenAI: GPT-5.1-Codex-Mini	openai	80%
23	Qwen: Qwen3 VL 32B Instruct	qwen	80%
24	Z.ai: GLM 4.6V	z-ai	80%
25	Google: Gemma 4 31B	google	80%
26	Baidu: ERNIE 4.5 VL 424B A47B	baidu	80%
27	Mistral: Mistral Large 3 2512	mistralai	80%
28	Google: Gemma 4 26B A4B	google	80%
29	ByteDance Seed: Seed-2.0-Lite	bytedance-seed	80%
30	Qwen: Qwen3.6 35B A3B	qwen	80%
31	Z.ai: GLM 5.2	z-ai	78%
32	Xiaomi: MiMo-V2.5-Pro	xiaomi	78%
33	Qwen: Qwen3.5-Flash	qwen	78%
34	MoonshotAI: Kimi K2.6	moonshotai	78%
35	MoonshotAI: Kimi K2.7 Code	moonshotai	78%
36	ByteDance Seed: Seed 1.6	bytedance-seed	78%
37	Qwen: Qwen3 VL 8B Thinking	qwen	78%
38	Meta: Llama 4 Scout	meta-llama	78%
39	Google: Gemini 4 Argon	google	76%
40	MoonshotAI: Kimi K3	moonshotai	76%
41	Mistral: Mistral Small 4	mistralai	76%
42	MiniMax: MiniMax-01	minimax	76%
43	Qwen: Qwen3.5-9B	qwen	76%
44	Qwen: Qwen3.5-35B-A3B	qwen	76%
45	Anthropic: Claude Fable 5	anthropic	73%
46	NVIDIA: Nemotron 3 Ultra	nvidia	73%
47	OpenAI: GPT-5.4 Nano	openai	73%
48	OpenAI: GPT-4.1 Mini	openai	73%
49	Qwen: Qwen2.5 VL 72B Instruct	qwen	73%
50	OpenAI: GPT-5.5 Pro	openai	71%
51	Qwen: Qwen3 VL 30B A3B Thinking	qwen	71%
52	Google: Gemini 3.1 Flash Lite Preview	google	71%
53	Qwen: Qwen3 VL 8B Instruct	qwen	71%
54	Qwen: Qwen3.5-27B	qwen	71%
55	Mistral: Ministral 3 3B 2512	mistralai	71%
56	Anthropic: Claude Opus 5.5	anthropic	68%
57	OpenAI: GPT-5 Mini	openai	68%
58	Google: Gemma 3 4B	google	68%
59	NVIDIA: Nemotron Nano 12B 2 VL (free)	nvidia	68%
60	Arcee AI: Trinity Large Thinking	arcee-ai	68%
61	Amazon: Nova Lite 1.0	amazon	68%
62	Qwen: Qwen3.5 Plus 2026-02-15	qwen	66%
63	Mistral: Ministral 3 8B 2512	mistralai	66%
64	Meta: Llama 4 Maverick	meta-llama	66%
65	Google: Gemma 3 12B	google	66%
66	Perplexity: Sonar	perplexity	66%
67	ByteDance Seed: Seed-2.0-Mini	bytedance-seed	66%
68	SpaceXAI: Grok 4.7	x-ai	63%
69	SpaceXAI: Grok 4.3	x-ai	63%
70	Qwen: Qwen3 VL 235B A22B Instruct	qwen	63%
71	ByteDance Seed: Seed 1.6 Flash	bytedance-seed	63%
72	Google: Gemini 2.5 Flash Lite	google	63%
73	OpenAI: GPT-5 Nano	openai	63%
74	Mistral: Mistral Small 3.2 24B	mistralai	63%
75	OpenAI: GPT-4o-mini	openai	63%
76	Google: Gemma 3 27B	google	61%
77	NVIDIA: Nemotron 3 Nano Omni (free)	nvidia	61%
78	OpenAI: GPT-4.1 Nano	openai	61%
79	OpenAI: GPT-6.1 Sol	openai	59%
80	StepFun: Step 3.7 Flash	stepfun	56%
81	Anthropic: Claude 3 Haiku	anthropic	54%
82	Meta: Llama 3.2 11B Vision Instruct	meta-llama	54%

← All rankings Methodology & 𝕍 →