MMODELYST
Evals/Humanity's Last Exam
EVL

Humanity's Last Exam

benchmarkReasoningaccuracy %
Leaderboard · 100 models
accuracy % — higher is better
1Claude Fable 5
53.3
2GPT-5.6 Sol (max)
47.2
3Claude Opus 4.8
45.7
4Muse Spark 1.1
45.1
5Gemini 3.1 Pro Preview
44.7
6GPT-5.6 Sol
44.7
7GPT-5.5
44.3
8GPT-5.6 Terra (max)
41.8
9GPT-5.4
41.6
10Gemini 3.5 Flash
41
11Grok 4.5
40.3
12GLM-5.2 (max)
40.1
13GPT-5.6 Terra
40
14Muse Spark
39.9
15GPT-5.3 Codex
39.9
16Claude Opus 4.7
39.6
17Claude Sonnet 5
39.6
18Qwen3.7 Max
38.1
19Gemini 3 Pro Preview
37.2
20GPT-5.6 Luna (max)
37.2
21MiniMax-M3
37.1
22Claude Opus 4.6
36.7
23Grok Build 0.1 0616
36
24Kimi K2.6
35.9
25DeepSeek V4 Pro
35.9
26GPT-5.6 Luna
35.6
27GPT-5.2
35.4
28Grok 4.3
35
29Gemini 3 Flash Preview
34.7
30MiMo-V2.5-Pro
33.8
31GPT-5.2 Codex
33.5
32Qwen3.7 Plus
33.4
33KAT-Coder-Pro V1
33.4
34Kimi K2.7 Code
32.8
35Nex-N2-Pro
32.4
36Grok 4.20 0309 v2
32.2
37DeepSeek V4 Flash
32.1
38Grok 4.20 0309
30
39Claude Sonnet 4.6
30
40Kimi K2.5
29.4
41Qwen3.6 Max Preview
28.9
42Claude Opus 4.5
28.4
43MiMo-V2-Pro
28.3
44MiniMax-M2.7
28.1
45GLM-5.1
28
46Qwen3.5 397B A17B
27.3
47GLM-5
27.2
48Nemotron 3 Ultra 550B A55B
26.6
49GPT-5.4 mini
26.6
50GPT-5.1
26.5
51GPT-5
26.5
52GPT-5.4 nano
26.5
53Qwen3 Max Thinking
26.2
54DeepSeek V3.2 Speciale
26.1
55Qwen3.6 Plus
25.7
56GPT-5 Codex
25.6
57Hy3-preview
25.5
58GLM-5-Turbo
25.4
59MiMo-V2.5
25.2
60GLM-4.7
25.1
61Grok 4
23.9
62GPT-5.1 Codex
23.4
63Qwen3.5 122B A10B
23.4
64Gemma 4 31B
22.7
65Step 3.5 Flash 2603
22.6
66Kimi K2 Thinking
22.3
67Qwen3.5 27B
22.2
68MiniMax-M2.1
22.2
69DeepSeek V3.2
22.2
70Qwen3.6 27B
21.6
71MiMo-V2-Flash
21.1
72Gemini 2.5 Pro
21.1
73MiMo-V2-Omni-0327
20.4
74GPT-5.5 Instant (May 2026)
20.3
75Qwen3.6 35B A3B
20.2
76MiMo-V2-Flash (Feb 2026)
20
77o3
20
78MiMo-V2-Omni
19.9
79Step 3.7 Flash
19.9
80Qwen3.5 35B A3B
19.7
81NVIDIA Nemotron 3 Super 120B A12B
19.2
82Step 3.5 Flash
19.1
83MiniMax-M2.5
19.1
84GPT-5.5 Instant (May 2026)
18.6
85gpt-oss-120b
18.5
86Ring-2.6-1T
18.3
87Gemma 4 26B A4B
18.3
88Grok 4.1 Fast
17.6
89o4-mini
17.5
90Claude 4.5 Sonnet
17.3
91Gemini 2.5 Pro Preview (Mar' 25)
17.1
92Grok 4 Fast
17
93GPT-5.1 Codex mini
16.9
94Gemini 3.1 Flash-Lite
16.2
95JT-4.1 Flash 236B A21B
16.1
96KAT Coder Pro V2
16
97GLM 5V Turbo
15.8
98Mercury 2
15.5
99Gemini 2.5 Pro Preview (May' 25)
15.4
100DeepSeek V3.1 Terminus
15.2
Benchmark health
headroom

Plenty of room at the top — this benchmark still separates frontier models clearly.

44.9
top-10 mean
53.3
best
100
models
Recent changes
GLM-5.2+8.46d ago
JT-4.1 Flash 236B A21B+16.16d ago
Grok 4.5+40.36d ago
Muse Spark 1.1+45.16d ago
GPT-5.6 Luna (max)+37.26d ago
What improves this score

Datasets & environments shown to raise it.

No data yet.
Research using this benchmark
all →
Trust but Verify: Prover-Verifier Deliberation for Selective LLM PredictionContext Training with Active Information SeekingThe Moltbook Files: A Harmless Slopocalypse or Humanity's Last ExperimentOpenSeeker-v2: Pushing the Limits of Search Agents with Informative and High-Difficulty Trajectories

Matched by name in title/abstract.