Skip to main content
benchlm.ai

Artificial Analysis Omniscience Hallucination Rate (AA-Omniscience Hallucination Rate)

A display-only Artificial Analysis factuality metric for the rate of incorrect answers among non-correct responses.

Benchmark score on AA-Omniscience Hallucination Rate — June 2, 2026

BenchLM mirrors the published score view for AA-Omniscience Hallucination Rate. Command A+ leads the public snapshot at 14.1% , followed by Qwen3.7 Max (22.9%) and MiMo-V2.5-Pro (24.5%). BenchLM does not use these results to rank models overall.

1Open

Command A+

Cohere

14.1%
Overall —Context 128K
2Closed

Qwen3.7 Max

Alibaba

22.9%
Overall 91Context 1M
3Closed

MiMo-V2.5-Pro

Xiaomi

24.5%
Overall —Context 1M
115 modelsKnowledgeCurrentDisplay onlyUpdated June 2, 2026

The published AA-Omniscience Hallucination Rate snapshot is tightly clustered at the top: Command A+ sits at 14.1%, while the third row is only 10.4 points behind. The broader top-10 spread is 19.9 points, so the benchmark still separates strong models even when the leaders cluster.

115 models have been evaluated on AA-Omniscience Hallucination Rate. The benchmark falls in the Knowledge category. This category carries a 12% weight in BenchLM.ai's overall scoring system. AA-Omniscience Hallucination Rate is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About AA-Omniscience Hallucination Rate

Year

2026

Tasks

Knowledge questions

Format

Hallucination rate

Difficulty

Factuality

BenchLM marks this row lower-is-better because a lower hallucination rate is preferable, even though the OpenRouter card displays the raw percentage.

Artificial Analysis model benchmarks

BenchLM freshness & provenance

Version

AA-Omniscience Hallucination Rate 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Benchmark score table (115 models)

1
Command A+
14.1%
2
Qwen3.7 Max
22.9%
3
MiMo-V2.5-Pro
24.5%
4
Grok 4.3
25.0%
5
GLM-5.1
29.4%
6
MiMo-V2-Pro
29.9%
7
Gemma 4 E4B
31.3%
8
Qwen3.6 Plus
32.0%
9
Gemma 4 E2B
32.9%
10
GLM-5
34.0%
11
MiniMax M2.7
34.4%
12
Claude Opus 4.8
35.9%
13
Claude Opus 4.7 (Adaptive)
36.2%
14
GPT-4o
37.9%
15
Kimi K2.6
39.3%
16
Claude 4 Sonnet
40.8%
17
Qwen 3.6 Max (preview)
44.2%
18
MiMo-V2-Omni
44.4%
19
Qwen3.6-27B
48.3%
20
Qwen3.6-35B-A3B
49.7%
21
Gemini 3.1 Pro
49.9%
22
Llama 3.1 405B
51.0%
23
GPT-5.1
51.3%
24
Claude Opus 4.7
51.9%
25
Claude Opus 4.5 Thinking
59.8%
26
Gemini 3.5 Flash
60.7%
27
Mistral Medium 3
60.9%
28
Claude Opus 4.6 (Adaptive)
61.3%
29
GLM-5-Turbo
62.2%
30
Grok 4
64.2%
31
Kimi K2.5 (Reasoning)
64.6%
32
Kimi K2.5
64.6%
33
Claude Sonnet 4.6
65.9%
34
Grok 4 Fast (Reasoning)
66.0%
35
GLM-4.6
66.1%
36
Mistral Small 4 (Reasoning)
66.8%
37
Mistral Small 4
66.8%
38
Mistral Large 2
67.8%
39
GLM-5V-Turbo
67.9%
40
o1
69.3%
41
Grok 4.1 Fast (Reasoning)
72.4%
42
GPT-5.2-Codex
72.8%
43
Muse Spark
73.2%
44
GPT-5.4 nano
73.6%
45
Kimi K2
74.2%
46
GPT-5.1-Codex-Max
74.4%
47
GPT-5.1-Codex
74.4%
48
MiMo-V2-Flash
75.1%
49
Claude Opus 4.5
75.4%
50
Claude Opus 4.6
76.0%
51
Granite-4.0-350M
77.8%
52
Nova Pro
77.9%
53
Claude 3 Haiku
78.2%
54
Llama 4 Scout
78.3%
55
Grok Code Fast 1
78.5%
56
GPT-4.1
79.6%
57
GPT-5.2
79.7%
58
Qwen3.5-27B
79.7%
59
Qwen3.5 397B
79.8%
60
GPT-5 (medium)
80.1%
61
DeepSeek V3.1 (Reasoning)
80.3%
62
GPT-4.1 nano
80.4%
63
Phi-4
80.5%
64
Gemma 4 26B A4B
80.9%
65
Exaone 4.0 32B
81.0%
66
Gemini 3.1 Flash-Lite
81.6%
67
Gemma 4 31B
81.6%
68
Nemotron Ultra 253B
81.7%
69
Grok 4.1 Fast
81.8%
70
GPT-4.1 mini
82.0%
71
Mistral Medium 3.5 128B
82.0%
72
GPT-5 (high)
82.1%
73
Nemotron 3 Nano Omni 30B A3B
83.1%
74
Granite-4.0-H-1B
83.4%
75
DeepSeek V3.1
83.5%
76
Mistral Large 3
83.7%
77
Qwen3.5-35B-A3B
84.0%
78
DeepSeek-R1
84.0%
79
Step 3.7 Flash
84.4%
80
GPT-5.5
85.5%
81
Qwen3.5-122B-A10B
85.5%
82
Trinity-Large-Preview
86.6%
83
Trinity-Large-Thinking
86.6%
84
GPT-5.3 Codex
86.9%
85
Hy3 Preview
86.9%
86
o3
87.1%
87
Llama 4 Maverick
87.3%
88
Gemini 2.5 Pro
87.4%
89
GPT-5.4
88.6%
90
DeepSeek V4 Pro (High)
88.6%
91
Qwen3.5 397B (Reasoning)
89.1%
92
K-Exaone
89.1%
93
DeepSeek V3
89.4%
94
Qwen3 Max
89.4%
95
Gemma 3 27B
89.5%
96
DeepSeek V4 Flash (High)
89.7%
97
GPT-5.4 mini
89.8%
98
Gemini 3 Flash
90.2%
99
GLM-4.7
90.3%
100
Gemini 3 Pro
90.9%
101
Nemotron 3 Nano 30B
90.9%
102
GPT-OSS 120B
91.2%
103
Solar Pro 2
91.5%
104
Exaone 4.0 1.2B
91.5%
105
GLM-4.5-Air
92.3%
106
Gemini 2.5 Flash
93.3%
107
DeepSeek V3.2
93.5%
108
Sarvam 105B
93.5%
109
Granite-4.0-1B
93.5%
110
DeepSeek V4 Pro (Max)
94.0%
111
GPT-OSS 20B
94.1%
112
Granite-4.0-H-350M
94.4%
113
DeepSeek V4 Flash (Max)
95.8%
114
Ling 2.6 Flash
95.8%
115
Sarvam 30B
97.0%

FAQ

What does AA-Omniscience Hallucination Rate measure?

A display-only Artificial Analysis factuality metric for the rate of incorrect answers among non-correct responses.

Which model scores highest on AA-Omniscience Hallucination Rate?

Command A+ by Cohere currently leads with a score of 14.1% on AA-Omniscience Hallucination Rate.

How many models are evaluated on AA-Omniscience Hallucination Rate?

115 AI models have been evaluated on AA-Omniscience Hallucination Rate on BenchLM.

Compare Top Models on AA-Omniscience Hallucination Rate

Command A+ vs Qwen3.7 MaxQwen3.7 Max vs MiMo-V2.5-ProMiMo-V2.5-Pro vs Grok 4.3Grok 4.3 vs GLM-5.1
Last updated: June 2, 2026 · BenchLM version AA-Omniscience Hallucination Rate 2026

The AI models change fast. We track them for you.

For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.