Hallucination

28 scenarios across 3 subcategories. 91 models scored.

Subcategories

Subcategory Avg Score Best Model Best Score
False Positives 49.53% o4 Mini High 97.19%
Content Invention 95.73% Claude Opus 4.6 (Reasoning) 100.00%
Output Corruption 98.79% Claude Opus 4.6 (Reasoning) 100.00%

Model Leaderboard

All models ranked by their Hallucination category score.

# Model Hallucination False Positives Content Invention Output Corruption Overall
1 Claude 3.5 Haiku 100.00% 100.00% 83.73%
2 Stealth: Aurora Alpha 99.93% 99.93% 83.79%
3 o4 Mini High 99.06% 97.19% 100.00% 100.00% 90.29%
4 Grok 4.1 Fast 99.02% 97.08% 100.00% 99.99% 89.55%
5 o4 Mini 98.75% 96.25% 100.00% 100.00% 88.35%
6 GPT-5.1 98.44% 95.31% 100.00% 100.00% 92.54%
7 Claude Opus 4.6 (Reasoning) 98.13% 94.38% 100.00% 100.00% 95.02%
8 Z.AI GLM 5 97.74% 93.23% 100.00% 100.00% 91.23%
9 GPT-5 Mini 97.71% 93.12% 100.00% 100.00% 92.62%
10 Gemini 2.5 Flash (Reasoning) 95.60% 86.87% 100.00% 99.94% 86.51%
11 Gemini 2.5 Flash Lite (Reasoning) 95.59% 91.56% 95.24% 99.96% 85.75%
12 GPT-4.1 95.24% 85.76% 100.00% 99.97% 88.68%
13 ByteDance Seed 1.6 95.14% 85.42% 100.00% 100.00% 90.70%
14 GPT-5.2 95.09% 85.28% 100.00% 100.00% 90.26%
15 ByteDance Seed 1.6 Flash 94.52% 94.11% 89.88% 99.57% 73.27%
16 Claude Sonnet 4.6 (Reasoning) 93.96% 81.88% 100.00% 100.00% 93.66%
17 Claude Opus 4.6 93.60% 80.81% 100.00% 100.00% 92.35%
18 Aion 2.0 93.57% 87.81% 92.89% 100.00% 89.21%
19 GPT-5 Nano 93.53% 92.50% 88.10% 99.98% 82.60%
20 Minimax M2.5 92.94% 79.32% 100.00% 99.50% 88.71%
21 Ministral 3 8B 92.52% 77.55% 100.00% 100.00% 71.76%
22 GPT-5 91.84% 75.52% 100.00% 100.00% 91.93%
23 Arcee AI: Trinity Mini 91.12% 73.37% 100.00% 100.00% 70.90%
24 Grok 4 Fast 91.09% 73.28% 100.00% 100.00% 86.15%
25 LFM2 24B 90.53% 71.58% 100.00% 100.00% 58.77%
26 Z.AI GLM 4.6 90.08% 71.35% 100.00% 98.89% 89.11%
27 Claude Sonnet 4.6 89.99% 69.98% 100.00% 100.00% 91.15%
28 Z.AI GLM 4.7 Flash 89.86% 76.72% 92.86% 100.00% 84.82%
29 Grok 4 89.45% 68.35% 100.00% 100.00% 88.12%
30 Ministral 8B 89.19% 67.57% 100.00% 100.00% 64.87%
31 Gemini 3.1 Pro (Preview) 89.06% 67.19% 100.00% 100.00% 94.37%
32 Z.AI GLM 4.7 88.47% 65.47% 100.00% 99.96% 88.69%
33 Gemini 3 Pro (Preview) 88.23% 64.69% 100.00% 100.00% 88.79%
34 MoonshotAI: Kimi K2.5 88.02% 64.06% 100.00% 100.00% 91.04%
35 GPT-4.1 Nano 87.73% 63.25% 100.00% 99.95% 71.94%
36 Z.AI GLM 4.5 87.05% 62.31% 100.00% 98.85% 86.27%
37 Qwen 3.5 122B 86.58% 59.74% 100.00% 100.00% 91.53%
38 Qwen 3.5 27B 86.58% 59.73% 100.00% 100.00% 90.85%
39 Gemini 2.5 Pro 86.11% 58.32% 100.00% 100.00% 88.53%
40 Gemini 3 Flash (Preview, Reasoning) 85.32% 55.98% 100.00% 99.97% 90.50%
41 Claude Haiku 4.5 83.86% 51.59% 100.00% 100.00% 85.14%
42 Qwen 3.5 397B A17B 82.10% 46.35% 100.00% 99.93% 91.73%
43 Mistral Medium 3.1 82.09% 46.27% 100.00% 100.00% 77.83%
44 Claude Opus 4.5 82.06% 46.19% 100.00% 100.00% 89.69%
45 GPT-4.1 Mini 81.14% 43.42% 100.00% 100.00% 83.20%
46 Qwen 3.5 35B 80.87% 49.74% 92.86% 100.00% 88.00%
47 Qwen 3.5 Flash 80.63% 49.01% 92.90% 99.98% 86.38%
48 Claude Sonnet 4 80.08% 40.23% 100.00% 100.00% 88.72%
49 Ministral 3 14B 79.99% 39.97% 100.00% 100.00% 72.54%
50 Hermes 3 405B 79.70% 39.09% 100.00% 100.00% 82.86%
51 Qwen 2.5 72B 79.56% 40.41% 100.00% 98.28% 75.46%
52 GPT-4o, Aug. 6th (temp=1) 79.53% 38.76% 100.00% 99.82% 82.62%
53 Mistral Large 3 78.17% 34.52% 100.00% 100.00% 85.43%
54 Mistral Large 2 77.87% 33.60% 100.00% 100.00% 82.41%
55 Mistral Large 77.50% 32.50% 100.00% 100.00% 80.15%
56 GPT-4o Mini (temp=1) 76.78% 30.35% 100.00% 100.00% 79.08%
57 Arcee AI: Trinity Large (Preview) 76.47% 29.41% 100.00% 100.00% 73.33%
58 Claude 3.5 Sonnet 76.31% 28.94% 100.00% 100.00% 84.24%
59 Gemini 2.5 Flash Lite 76.17% 28.56% 100.00% 99.96% 81.08%
60 Mistral Small 3.2 24B 75.83% 28.78% 100.00% 98.73% 78.60%
61 Claude Opus 4 75.68% 27.04% 100.00% 100.00% 87.69%
62 Claude Sonnet 4.5 75.57% 26.70% 100.00% 100.00% 88.03%
63 Claude 3.7 Sonnet 75.18% 25.54% 100.00% 100.00% 83.39%
64 Llama 3.1 Nemotron 70B 74.99% 28.79% 96.43% 99.75% 74.70%
65 Mistral Small Creative 74.46% 23.38% 100.00% 100.00% 73.27%
66 GPT-4o Mini (temp=0) 73.56% 20.67% 100.00% 100.00% 78.29%
67 GPT-4o, May 13th (temp=1) 73.40% 20.21% 100.00% 100.00% 83.80%
68 GPT-4o, Aug. 6th (temp=0) 73.35% 20.06% 100.00% 100.00% 82.45%
69 Qwen 3.5 Plus (2026-02-15) 73.35% 20.40% 100.00% 99.65% 85.96%
70 DeepSeek V3 (2024-12-26) 73.11% 19.33% 100.00% 100.00% 83.68%
71 DeepSeek V3.1 72.80% 18.52% 100.00% 99.89% 82.39%
72 DeepSeek V3.2 72.50% 17.51% 100.00% 100.00% 82.25%
73 Writer: Palmyra X5 72.04% 23.27% 92.86% 100.00% 79.57%
74 Gemini 2.5 Flash 71.70% 16.03% 100.00% 99.08% 80.60%
75 Gemini 3 Flash (Preview) 71.24% 13.77% 100.00% 99.95% 85.35%
76 Ministral 3B 70.75% 26.52% 85.71% 100.00% 61.29%
77 Ministral 3 3B 70.45% 22.08% 89.29% 100.00% 67.22%
78 WizardLM 2 8x22b 70.24% 29.74% 85.73% 95.25% 71.07%
79 Llama 3.1 70B 69.78% 29.26% 92.86% 87.24% 78.40%
80 GPT-4o, May 13th (temp=0) 69.76% 9.29% 100.00% 100.00% 85.36%
81 DeepSeek-V2 Chat 69.48% 17.97% 90.48% 100.00% 84.83%
82 Gemma 3 12B 69.15% 7.45% 100.00% 100.00% 78.41%
83 Gemma 3 27B 68.74% 6.23% 100.00% 100.00% 77.85%
84 Gemma 3 4B 67.60% 2.80% 100.00% 100.00% 68.57%
85 DeepSeek V3 (2025-03-24) 67.07% 18.55% 88.10% 94.58% 81.99%
86 Hermes 3 70B 67.05% 42.95% 60.79% 97.41% 72.57%
87 Cohere Command R+ (Aug. 2024) 64.40% 55.29% 37.90% 100.00% 69.03%
88 Mistral NeMO 62.63% 11.73% 76.19% 99.97% 65.04%
89 Claude 3 Haiku 60.81% 19.94% 62.50% 100.00% 71.19%
90 Rocinante 12B 52.09% 27.40% 38.95% 89.92% 54.55%
91 Llama 3.1 8B 43.41% 18.43% 77.38% 34.42% 63.37%