Skip to main content
benchlm.ai

Artificial Analysis SciCode (AA-SciCode)

A display-only Artificial Analysis SciCode score.

Benchmark score on AA-SciCode — May 21, 2026

BenchLM mirrors the published score view for AA-SciCode. Gemini 3.1 Pro leads the public snapshot at 58.9% , followed by GPT-5.4 (56.6%) and GPT-5.5 (56.1%). BenchLM does not use these results to rank models overall.

1Closed

Gemini 3.1 Pro

Google

58.9%
Overall 92Context 1M
2Closed

GPT-5.4

OpenAI

56.6%
Overall 89Context 1.05M
3Closed

GPT-5.5

OpenAI

56.1%
Overall 91Context 1M
125 modelsCodingCurrentDisplay onlyUpdated May 21, 2026

The published AA-SciCode snapshot is tightly clustered at the top: Gemini 3.1 Pro sits at 58.9%, while the third row is only 2.8 points behind. The broader top-10 spread is 6.8 points, so many of the published scores sit in a relatively narrow band.

125 models have been evaluated on AA-SciCode. The benchmark falls in the Coding category. This category carries a 20% weight in BenchLM.ai's overall scoring system. AA-SciCode is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About AA-SciCode

Year

2026

Tasks

Scientific coding subproblems

Format

Task success rate

Difficulty

Scientific programming

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

Artificial Analysis SciCode Benchmark Leaderboard

BenchLM freshness & provenance

Version

AA-SciCode 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
58.9%
2
GPT-5.4
56.6%
3
GPT-5.5
56.1%
4
Gemini 3 Pro
56.1%
5
GPT-5.2-Codex
54.6%
6
Claude Opus 4.7 (Adaptive)
54.5%
7
Kimi K2.6
53.5%
8
GPT-5.3 Codex
53.2%
9
Gemini 3.5 Flash
53.1%
10
GPT-5.2
52.1%
11
Claude Opus 4.6 (Adaptive)
51.9%
12
Muse Spark
51.5%
13
MiMo-V2.5-Pro
50.2%
14
Claude Opus 4.7
50.1%
15
DeepSeek V4 Pro (Max)
50.0%
16
Gemini 3 Flash
49.9%
17
GPT-5.4 mini
49.9%
18
Claude Opus 4.5 Thinking
49.5%
19
Kimi K2.5 (Reasoning)
49.0%
20
Kimi K2.5
49.0%
21
Qwen3.7 Max
48.8%
22
Grok 4.3
47.3%
23
Claude Opus 4.5
47.0%
24
MiniMax M2.7
47.0%
25
Claude Sonnet 4.6
46.9%
26
Qwen 3.6 Max (preview)
46.9%
27
GPT-5.4 nano
46.9%
28
DeepSeek V4 Pro (High)
46.4%
29
GLM-5 (Reasoning)
46.2%
30
GLM-5
46.2%
31
Claude Opus 4.6
45.7%
32
Grok 4
45.7%
33
GLM-4.7
45.1%
34
DeepSeek V4 Flash (Max)
44.9%
35
Grok 4.1 Fast (Reasoning)
44.2%
36
Grok 4 Fast (Reasoning)
44.2%
37
GLM-5.1
43.8%
38
GLM-5-Turbo
43.6%
39
GLM-5V-Turbo
43.5%
40
Gemma 4 31B
43.4%
41
GPT-5.1
43.3%
42
GPT-5 (high)
42.9%
43
Gemini 2.5 Pro
42.8%
44
MiMo-V2-Pro
42.5%
45
Qwen3.5 397B (Reasoning)
42.0%
46
DeepSeek V4 Flash (High)
42.0%
47
Qwen3.5-122B-A10B
42.0%
48
Gemini 3.1 Flash-Lite
41.9%
49
Hy3 Preview
41.2%
50
GPT-5 (medium)
41.1%
51
Qwen3.5 397B
41.1%
52
o3
41.0%
53
Claude 4.1 Opus Thinking
40.9%
54
Qwen3.6 Plus
40.7%
55
Grok 3 Mini
40.6%
56
GPT-4.1 mini
40.4%
57
DeepSeek-R1
40.3%
58
GPT-5.1-Codex-Max
40.2%
59
GPT-5.1-Codex
40.2%
60
Gemma 4 26B A4B
40.0%
61
o3-mini
39.9%
62
Qwen3.6-27B
39.8%
63
Mistral Medium 3.5 128B
39.6%
64
Qwen3.5-27B
39.5%
65
DeepSeek V3.1 (Reasoning)
39.1%
66
GPT-OSS 120B
38.9%
67
DeepSeek V3.2
38.7%
68
Qwen3 Max
38.3%
69
GPT-4.1
38.1%
70
Mistral Small 4 (Reasoning)
38.0%
71
Mistral Small 4
38.0%
72
Command A+
37.8%
73
Qwen3.5-35B-A3B
37.7%
74
DeepSeek R1 Distill Qwen 32B
37.6%
75
Claude 4 Sonnet
37.3%
76
DeepSeek V3.1
36.7%
77
MiMo-V2-Omni
36.7%
78
Mistral Large 3
36.2%
79
Grok Code Fast 1
36.2%
80
Trinity-Large-Preview
36.1%
81
Trinity-Large-Thinking
36.1%
82
Qwen3.6-35B-A3B
35.8%
83
o1
35.8%
84
K-Exaone
35.6%
85
DeepSeek V3
35.4%
86
Nemotron Ultra 253B
34.7%
87
Kimi K2
34.5%
88
GPT-OSS 20B
34.4%
89
GPT-4o
33.3%
90
Llama 4 Maverick
33.1%
91
Mistral Medium 3
33.1%
92
GLM-4.6
33.1%
93
Grok 4.20
32.8%
94
GPT-4 Turbo
31.9%
95
GLM-4.5-Air
30.6%
96
Llama 3.1 405B
29.9%
97
Grok 4.1 Fast
29.6%
98
Gemini 1.5 Pro
29.5%
99
Mistral Large 2
29.2%
100
Gemini 2.5 Flash
29.1%
101
Nemotron 3 Nano Omni 30B A3B
27.8%
102
Ling 2.6 Flash
27.1%
103
Qwen2.5 Coder 32B Instruct
27.1%
104
Sarvam 105B
26.4%
105
Phi-4
26.0%
106
MiMo-V2-Flash
25.9%
107
GPT-4.1 nano
25.9%
108
Exaone 4.0 32B
25.2%
109
Solar Pro 2
24.8%
110
Gemma 4 E4B
24.4%
111
Claude 3 Opus
23.3%
112
Nemotron 3 Nano 30B
23.0%
113
GPT-4o mini
22.9%
114
Gemma 3 27B
21.2%
115
Gemma 4 E2B
20.9%
116
Nova Pro
20.8%
117
Sarvam 30B
19.2%
118
Claude 3 Haiku
18.6%
119
Llama 4 Scout
17.0%
120
Gemini 1.0 Pro
11.7%
121
Granite-4.0-1B
8.7%
122
Granite-4.0-H-1B
8.2%
123
Exaone 4.0 1.2B
7.4%
124
Granite-4.0-H-350M
1.7%
125
Granite-4.0-350M
0.9%

FAQ

What does AA-SciCode measure?

A display-only Artificial Analysis SciCode score.

Which model scores highest on AA-SciCode?

Gemini 3.1 Pro by Google currently leads with a score of 58.9% on AA-SciCode.

How many models are evaluated on AA-SciCode?

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

Compare Top Models on AA-SciCode

Gemini 3.1 Pro vs GPT-5.4GPT-5.4 vs GPT-5.5GPT-5.5 vs Gemini 3 ProGemini 3 Pro vs GPT-5.2-Codex
Last updated: May 21, 2026 · BenchLM version AA-SciCode 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.