Skip to main content
Meo

Critical thinking (Watson–Glaser)

Inference, assumptions, deduction, interpretation, argument strength. As of 2026-10-18, the strongest model on this domain is OpenAI: GPT-5.5 (100% — 4 of 4 questions correct).

What this domain tests

Watson–Glaser-style critical-thinking appraisal: drawing valid inferences, recognizing unstated assumptions, deducing conclusions, interpreting evidence, and evaluating argument strength.

Why it resists contamination

Fresh items on a private holdout graded by the multi-lab jury — the appraisal structure is public but the specific passages and keys are not. See the full methodology.

Best AI models for Critical thinking (Watson–Glaser)

#ModelCritical thinking (Watson–Glaser) accuracyQuestionsOverall𝕍 index
1OpenAI: GPT-5.5100%4/494.1%23.2
2Google: Gemini 4 Argon100%4/492.5%18.2
3Anthropic: Claude Opus 5.5100%4/492.1%100.0
4OpenAI: GPT-6.1 Sol100%4/490.4%35.6
5Anthropic: Claude Fable 5100%4/488.7%27.5
6Anthropic: Claude Opus 4.8100%4/484.9%12.1
7Google: Gemini 3.5 Flash100%4/482.0%11.1
8MoonshotAI: Kimi K3100%4/480.3%3.6
9Google: Gemini 3.1 Pro Preview100%4/479.5%3.2
10Qwen: Qwen3.7 Plus100%4/466.9%0.051
11Nex AGI: Nex-N2-Pro (free)100%4/465.7%0.18
12Z.ai: GLM 4.6V100%4/465.7%0.12
13Google: Gemini 3.1 Flash Lite Preview100%4/461.9%0.46
14Mistral: Mistral Large 3 2512100%4/459.0%0.17
15Google: Gemini 3.1 Flash Lite100%4/456.5%0.13
16Meta: Llama 4 Maverick100%4/452.7%0.008
17Google: Gemma 3 12B100%4/449.0%0.009
18Qwen: Qwen3.6 Plus100%4/444.8%<0.001
19OpenAI: GPT-5.5 Pro75%3/491.6%4.3
20Z.ai: GLM 5.275%3/483.2%0.78
21Qwen: Qwen3.7 Max75%3/481.0%0.65
22OpenAI: GPT-5.4 Nano75%3/469.5%2.3
23Qwen: Qwen3.5-Flash75%3/468.2%0.15
24Qwen: Qwen3 VL 30B A3B Instruct75%3/466.1%0.13
25Qwen: Qwen3 VL 30B A3B Thinking75%3/464.0%0.008
26Google: Gemma 4 31B75%3/463.6%0.009
27OpenAI: GPT-4.1 Mini75%3/463.6%0.26
28Baidu: ERNIE 4.5 VL 424B A47B 75%3/463.2%0.015
29Qwen: Qwen3.5 Plus 2026-02-1575%3/461.9%0.012
30Google: Gemini 2.5 Flash Lite75%3/457.3%0.13
31Google: Gemma 4 26B A4B 75%3/455.2%0.003
32Qwen: Qwen3.5-27B75%3/449.4%<0.001
33OpenAI: GPT-4.1 Nano75%3/448.1%0.044
34ByteDance Seed: Seed-2.0-Lite75%3/447.7%<0.001
35Qwen: Qwen3.5-35B-A3B75%3/439.3%<0.001
36Owl Alpha50%2/473.7%0.37
37Xiaomi: MiMo-V2.5-Pro50%2/471.5%0.18
38Qwen: Qwen3.6 Flash50%2/469.5%0.17
39MoonshotAI: Kimi K2.7 Code50%2/466.9%0.073
40OpenAI: GPT-5 Mini50%2/466.9%0.25
41Qwen: Qwen3 VL 32B Instruct50%2/466.1%0.19
42Z.ai: GLM 5.150%2/463.5%0.033
43ByteDance Seed: Seed 1.6 Flash50%2/457.7%0.080
44Mistral: Mistral Small 450%2/457.7%0.065
45Perceptron: Perceptron Mk150%2/457.7%0.055
46Mistral: Ministral 3 8B 251250%2/454.0%0.065
47Mistral: Mistral Small 3.2 24B50%2/453.6%0.017
48Qwen: Qwen2.5 VL 72B Instruct50%2/450.2%0.005
49MiniMax: MiniMax-0150%2/449.4%0.005
50OpenAI: GPT-4o-mini50%2/447.7%0.021
51Qwen: Qwen3.6 35B A3B50%2/447.7%0.004
52Google: Gemma 3 4B50%2/447.3%0.003
53Perplexity: Sonar50%2/445.6%0.004
54NVIDIA: Nemotron Nano 12B 2 VL (free)50%2/442.3%0.002
55Anthropic: Claude 3 Haiku50%2/441.8%0.014
56inclusionAI: Ring-2.6-1T25%1/484.7%14.8
57DeepSeek: DeepSeek V4 Flash 042325%1/481.8%0.43
58DeepSeek: DeepSeek V4 Pro 042325%1/480.3%0.53
59SpaceXAI: Grok 4.725%1/472.4%0.17
60MiniMax: MiniMax M325%1/467.4%0.13
61MoonshotAI: Kimi K2.625%1/466.9%0.050
62OpenAI: GPT-5.1-Codex-Mini25%1/466.5%1.3
63ByteDance Seed: Seed 1.625%1/464.9%0.036
64Qwen: Qwen3 VL 235B A22B Instruct25%1/464.4%0.060
65Qwen: Qwen3 VL 8B Instruct25%1/456.9%0.028
66Tencent: Hy3 preview25%1/454.0%0.007
67Google: Gemma 3 27B25%1/450.2%0.006
68StepFun: Step 3.7 Flash25%1/449.8%0.005
69Meta: Llama 4 Scout25%1/448.5%0.009
70Mistral: Ministral 3 3B 251225%1/445.6%0.021
71Arcee AI: Trinity Large Thinking25%1/441.6%<0.001
72Amazon: Nova Lite 1.025%1/439.7%0.012
73Meta: Llama 3.2 11B Vision Instruct25%1/435.6%<0.001
74NVIDIA: Nemotron 3 Ultra0%0/475.2%1.2
75SpaceXAI: Grok 4.30%0/466.5%1.7
76Xiaomi: MiMo-V2.50%0/466.5%0.26
77Qwen: Qwen3 VL 8B Thinking0%0/463.2%0.072
78Mistral: Ministral 3 14B 25120%0/456.5%0.086
79OpenAI: GPT-5 Nano0%0/454.0%0.026
80NVIDIA: Nemotron 3 Nano Omni (free)0%0/448.5%0.016
81Qwen: Qwen3.5-9B0%0/441.4%<0.001
82ByteDance Seed: Seed-2.0-Mini0%0/437.2%<0.001

Primary sources & further reading

Frequently asked questions

What does the meo Critical thinking (Watson–Glaser) benchmark test?

Watson–Glaser-style critical-thinking appraisal: drawing valid inferences, recognizing unstated assumptions, deducing conclusions, interpreting evidence, and evaluating argument strength.

Why can't AI models game the Critical thinking (Watson–Glaser) benchmark?

Fresh items on a private holdout graded by the multi-lab jury — the appraisal structure is public but the specific passages and keys are not.

Which AI model is best at Critical thinking (Watson–Glaser)?

OpenAI: GPT-5.5 leads the Critical thinking (Watson–Glaser) domain on the meo benchmark with 100% accuracy (as of 2026-10-18).

Related: all domains · Critical thinking (Watson–Glaser) ranking · methodology · full leaderboard.