Skip to main content
benchlm.ai

Artificial Analysis Humanity's Last Exam (AA-HLE)

A display-only Artificial Analysis Humanity's Last Exam score.

Benchmark score on AA-HLE — May 21, 2026

BenchLM mirrors the published score view for AA-HLE. Gemini 3.1 Pro leads the public snapshot at 44.7% , followed by GPT-5.5 (44.3%) and GPT-5.4 (41.6%). BenchLM does not use these results to rank models overall.

1Closed

Gemini 3.1 Pro

Google

44.7%
Overall 92Context 1M
2Closed

GPT-5.5

OpenAI

44.3%
Overall 91Context 1M
3Closed

GPT-5.4

OpenAI

41.6%
Overall 89Context 1.05M
125 modelsKnowledgeCurrentDisplay onlyUpdated May 21, 2026

The published AA-HLE snapshot is tightly clustered at the top: Gemini 3.1 Pro sits at 44.7%, while the third row is only 3.1 points behind. The broader top-10 spread is 8.0 points, so many of the published scores sit in a relatively narrow band.

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

About AA-HLE

Year

2026

Tasks

Expert-level questions

Format

Accuracy

Difficulty

Frontier expert reasoning

BenchLM stores the Artificial Analysis HLE result separately from the weighted HLE lane so AA refreshes remain display-only.

Artificial Analysis Humanity's Last Exam Benchmark Leaderboard

BenchLM freshness & provenance

Version

AA-HLE 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 (125 models)

1
Gemini 3.1 Pro
44.7%
2
GPT-5.5
44.3%
3
GPT-5.4
41.6%
4
Gemini 3.5 Flash
41.0%
5
GPT-5.3 Codex
39.9%
6
Muse Spark
39.9%
7
Claude Opus 4.7 (Adaptive)
39.6%
8
Qwen3.7 Max
38.1%
9
Gemini 3 Pro
37.2%
10
Claude Opus 4.6 (Adaptive)
36.7%
11
DeepSeek V4 Pro (Max)
35.9%
12
Kimi K2.6
35.9%
13
GPT-5.2
35.4%
14
Grok 4.3
35.0%
15
MiMo-V2.5-Pro
33.8%
16
DeepSeek V4 Pro (High)
33.5%
17
GPT-5.2-Codex
33.5%
18
DeepSeek V4 Flash (Max)
32.1%
19
Claude Opus 4.7
31.2%
20
Kimi K2.5 (Reasoning)
29.4%
21
Kimi K2.5
29.4%
22
Qwen 3.6 Max (preview)
28.9%
23
Claude Opus 4.5 Thinking
28.4%
24
MiMo-V2-Pro
28.3%
25
MiniMax M2.7
28.1%
26
GLM-5.1
28.0%
27
DeepSeek V4 Flash (High)
27.8%
28
Qwen3.5 397B (Reasoning)
27.3%
29
GLM-5 (Reasoning)
27.2%
30
GLM-5
27.2%
31
GPT-5.4 mini
26.6%
32
GPT-5.1
26.5%
33
GPT-5 (high)
26.5%
34
GPT-5.4 nano
26.5%
35
Qwen3.6 Plus
25.7%
36
Hy3 Preview
25.5%
37
GLM-5-Turbo
25.4%
38
GLM-4.7
25.1%
39
Grok 4.20
24.2%
40
Grok 4
23.9%
41
GPT-5 (medium)
23.5%
42
GPT-5.1-Codex-Max
23.4%
43
Qwen3.5-122B-A10B
23.4%
44
GPT-5.1-Codex
23.4%
45
Gemma 4 31B
22.7%
46
Qwen3.5-27B
22.2%
47
Qwen3.6-27B
21.6%
48
Gemini 2.5 Pro
21.1%
49
Qwen3.6-35B-A3B
20.2%
50
o3
20.0%
51
MiMo-V2-Omni
19.9%
52
Qwen3.5-35B-A3B
19.7%
53
Qwen3.5 397B
18.8%
54
Claude Opus 4.6
18.6%
55
GPT-OSS 120B
18.5%
56
Gemma 4 26B A4B
18.3%
57
Grok 4.1 Fast (Reasoning)
17.6%
58
Grok 4 Fast (Reasoning)
17.0%
59
Gemini 3.1 Flash-Lite
16.2%
60
GLM-5V-Turbo
15.8%
61
DeepSeek-R1
14.9%
62
Trinity-Large-Preview
14.7%
63
Trinity-Large-Thinking
14.7%
64
Gemini 3 Flash
14.1%
65
Claude Sonnet 4.6
13.2%
66
K-Exaone
13.1%
67
DeepSeek V3.1 (Reasoning)
13.0%
68
Claude Opus 4.5
12.9%
69
Mistral Medium 3.5 128B
12.8%
70
Claude 4.1 Opus Thinking
11.9%
71
Command A+
11.4%
72
Grok 3 Mini
11.1%
73
Qwen3 Max
11.1%
74
DeepSeek V3.2
10.5%
75
Sarvam 105B
10.1%
76
GPT-OSS 20B
9.8%
77
Mistral Small 4 (Reasoning)
9.5%
78
Mistral Small 4
9.5%
79
o3-mini
8.7%
80
Nemotron Ultra 253B
8.1%
81
MiMo-V2-Flash
8.0%
82
o1
7.7%
83
Grok Code Fast 1
7.5%
84
Kimi K2
7.0%
85
Sarvam 30B
7.0%
86
GLM-4.5-Air
6.8%
87
Granite-4.0-H-350M
6.4%
88
DeepSeek V3.1
6.3%
89
Ling 2.6 Flash
6.2%
90
Exaone 4.0 1.2B
5.8%
91
Granite-4.0-350M
5.7%
92
DeepSeek R1 Distill Qwen 32B
5.5%
93
Nemotron 3 Nano Omni 30B A3B
5.3%
94
GLM-4.6
5.2%
95
Gemini 2.5 Flash
5.1%
96
Granite-4.0-1B
5.1%
97
Grok 4.1 Fast
5.0%
98
Granite-4.0-H-1B
5.0%
99
Gemini 1.5 Pro
4.9%
100
Exaone 4.0 32B
4.9%
101
Llama 4 Maverick
4.8%
102
Gemma 4 E2B
4.8%
103
Gemma 3 27B
4.7%
104
GPT-4.1
4.6%
105
GPT-4.1 mini
4.6%
106
Nemotron 3 Nano 30B
4.6%
107
Gemini 1.0 Pro
4.6%
108
Llama 4 Scout
4.3%
109
Mistral Medium 3
4.3%
110
Llama 3.1 405B
4.2%
111
Mistral Large 3
4.1%
112
Phi-4
4.1%
113
GPT-4o mini
4.0%
114
Claude 4 Sonnet
4.0%
115
Mistral Large 2
4.0%
116
GPT-4.1 nano
3.9%
117
Claude 3 Haiku
3.9%
118
Qwen2.5 Coder 32B Instruct
3.8%
119
Solar Pro 2
3.8%
120
Gemma 4 E4B
3.7%
121
DeepSeek V3
3.6%
122
Nova Pro
3.4%
123
GPT-4o
3.3%
124
GPT-4 Turbo
3.3%
125
Claude 3 Opus
3.1%

FAQ

What does AA-HLE measure?

A display-only Artificial Analysis Humanity's Last Exam score.

Which model scores highest on AA-HLE?

Gemini 3.1 Pro by Google currently leads with a score of 44.7% on AA-HLE.

How many models are evaluated on AA-HLE?

125 AI models have been evaluated on AA-HLE on BenchLM.

Compare Top Models on AA-HLE

Gemini 3.1 Pro vs GPT-5.5GPT-5.5 vs GPT-5.4GPT-5.4 vs Gemini 3.5 FlashGemini 3.5 Flash vs GPT-5.3 Codex
Last updated: May 21, 2026 · BenchLM version AA-HLE 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.