Best AI Models for Engineering

Engineering: score by model

15 questions · the bar shows each score's 95% interval

Engineering leaderboard
#	Model	Correct	Score	95% interval
1	Gemini 4 ArgonGoogle	13 / 14	92.9%	69–99%
2	Claude Fable 5.1Anthropic · statistically tied with #1	13 / 15	86.7%	62–96%
2	Claude Sonnet 5.5Anthropic · statistically tied with #1	13 / 15	86.7%	62–96%
4	Claude Opus 5.5Anthropic · statistically tied with #1	12 / 15	80.0%	55–93%
4	Gemini 3.8 FlashGoogle · statistically tied with #1	12 / 15	80.0%	55–93%
6	Muse WonderMeta · statistically tied with #1	11 / 15	73.3%	48–89%
6	Grok 4.7SpaceXAI · statistically tied with #1	11 / 15	73.3%	48–89%
6	GPT-5.6 TerraOpenAI · statistically tied with #1	11 / 15	73.3%	48–89%
6	GPT-5.3-CodexOpenAI · statistically tied with #1	11 / 15	73.3%	48–89%
6	Muse Spark OpenMeta · statistically tied with #1	11 / 15	73.3%	48–89%
6	Kimi K3.1 ProMoonshotAI · statistically tied with #1	11 / 15	73.3%	48–89%
12	GPT-6 AstraOpenAI · statistically tied with #1	10 / 15	66.7%	42–85%
12	GPT-6.1 Terra ProOpenAI · statistically tied with #1	10 / 15	66.7%	42–85%
12	DeepSeek V4 Flash Vision ExpDeepSeek · statistically tied with #1	10 / 15	66.7%	42–85%
12	Gemini 3.5 Flash LiteGoogle · statistically tied with #1	10 / 15	66.7%	42–85%
12	Gemini 3.1 Pro PreviewGoogle · statistically tied with #1	10 / 15	66.7%	42–85%
12	GLM 5.3Z.ai · statistically tied with #1	10 / 15	66.7%	42–85%
12	GPT-6.1 SolOpenAI · statistically tied with #1	10 / 15	66.7%	42–85%
12	GPT-5.4OpenAI · statistically tied with #1	10 / 15	66.7%	42–85%
12	Claude Haiku 5.5Anthropic · statistically tied with #1	10 / 15	66.7%	42–85%
21	GPT-6.1 TerraOpenAI · statistically tied with #1	9 / 15	60.0%	36–80%
21	DeepSeek V4.1 Flash 1017DeepSeek · statistically tied with #1	9 / 15	60.0%	36–80%
21	GPT-6 LunaOpenAI · statistically tied with #1	9 / 15	60.0%	36–80%
21	Qwen3.8 Max (0902)Qwen · statistically tied with #1	9 / 15	60.0%	36–80%
25	o3OpenAI · statistically tied with #1	8 / 15	53.3%	30–75%
25	GLM 5.3 FlashZ.ai · statistically tied with #1	8 / 15	53.3%	30–75%
25	DeepSeek V4 Pro 0813DeepSeek · statistically tied with #1	8 / 15	53.3%	30–75%
25	GPT-6 SolOpenAI · statistically tied with #1	8 / 15	53.3%	30–75%
25	MiMo-V2.6-ProXiaomi · statistically tied with #1	8 / 15	53.3%	30–75%
30	Kimi K2.7 CodeMoonshotAI · statistically tied with #1	7 / 15	46.7%	25–70%
30	MiMo-V2.6-FlashXiaomi · statistically tied with #1	7 / 15	46.7%	25–70%
32	Llama 4 MaverickMeta	6 / 14	42.9%	21–67%
33	InklingThinking Machines	6 / 15	40.0%	20–64%
33	Step 5 PreviewStepFun	6 / 15	40.0%	20–64%
33	Mercury 2.5Inception	6 / 15	40.0%	20–64%
33	Phi 4Microsoft	6 / 15	40.0%	20–64%
33	Hy3Tencent	6 / 15	40.0%	20–64%
33	Qwen3 Coder NextQwen	6 / 15	40.0%	20–64%
33	Inkling SmallThinking Machines	6 / 15	40.0%	20–64%
33	Qwen3.8 27BQwen	6 / 15	40.0%	20–64%
33	Nemotron 3 UltraNVIDIA	6 / 15	40.0%	20–64%
33	MiniMax M3.1 FlashMiniMax	6 / 15	40.0%	20–64%
33	gpt-oss-120bOpenAI	6 / 15	40.0%	20–64%
44	Qwen3.8 2.4T A95BQwen	5 / 15	33.3%	15–58%
44	MiniMax M3MiniMax	5 / 15	33.3%	15–58%
46	Qwen3.7 PlusQwen	4 / 15	26.7%	11–52%
46	Nemotron 3.5 Super VLNVIDIA	4 / 15	26.7%	11–52%
46	gpt-oss-20bOpenAI	4 / 15	26.7%	11–52%
46	Ministral 3 8B 2512Mistral	4 / 15	26.7%	11–52%
46	Nemotron 3.5 LightningNVIDIA	4 / 15	26.7%	11–52%
46	Granite 4.2 8BIBM	4 / 15	26.7%	11–52%
52	Command ACohere	3 / 15	20.0%	7–45%
52	Mistral Large 4Mistral	3 / 15	20.0%	7–45%
54	Llama 4 ScoutMeta	2 / 15	13.3%	4–38%
54	Ministral 3 3B 2512Mistral	2 / 15	13.3%	4–38%
54	Solar Mini 4Upstage	2 / 15	13.3%	4–38%
54	Ministral 3 14B 2512Mistral	2 / 15	13.3%	4–38%
54	Llama 3.3 70B InstructMeta	2 / 15	13.3%	4–38%
54	Qwen3.6 35B A3BQwen	2 / 15	13.3%	4–38%
60	Ling 3.1 FlashinclusionAI	1 / 15	6.7%	1–30%
60	Granite 4.0 MicroIBM	1 / 15	6.7%	1–30%
60	Nova Micro 1.0Amazon	1 / 15	6.7%	1–30%
60	Solar Pro 4Upstage	1 / 15	6.7%	1–30%
60	LongCat 2.0Meituan	1 / 15	6.7%	1–30%
65	Mistral Medium 3.5Mistral	0 / 15	0.0%	0–20%
65	Reka Flash 3rekaai	0 / 15	0.0%	0–20%
65	Gemma 4 26B A4BGoogle	0 / 15	0.0%	0–20%
65	Gemma 4 31BGoogle	0 / 15	0.0%	0–20%
65	Qwen3.5 397B A17BQwen	0 / 15	0.0%	0–20%
65	Hermes 4 405BNous	0 / 15	0.0%	0–20%
65	Mistral Small 4Mistral	0 / 15	0.0%	0–20%
65	Trinity Large ThinkingArcee AI	0 / 15	0.0%	0–20%

How hard are these questions?

Questions by the share of 72 scored models that got them right

1No model right1 question
2Under 25% of models3 questions
325–50% of models5 questions
450–75% of models5 questions
575% or more, not all1 question
6Every model right0 questions

Engineering: frequently asked questions

Which AI model is best at engineering?

Gemini 4 Argon answered the most engineering questions correctly: 13 of 14. With only 15 questions, 30 other models are statistically tied with the leader, so treat this ranking as a rough guide.

Where do the engineering questions come from?

14 from SuperGPQA (ODC-BY 1.0) and 1 from MMLU-Pro (MIT). Every question is multiple choice with ten options and one correct answer.

How hard are the engineering questions?

1 of the 15 questions were answered correctly by none of the 72 scored models, and 0 by all of them.

Model

Correct

Score

95% interval

Gemini 4 ArgonGoogle

13 / 14

92.9%

69–99%

Claude Fable 5.1Anthropic · statistically tied with #1

13 / 15

86.7%

62–96%

Claude Sonnet 5.5Anthropic · statistically tied with #1

13 / 15

86.7%

62–96%

Claude Opus 5.5Anthropic · statistically tied with #1

12 / 15

80.0%

55–93%

Gemini 3.8 FlashGoogle · statistically tied with #1

12 / 15

80.0%

55–93%

Muse WonderMeta · statistically tied with #1

11 / 15

73.3%

48–89%

Grok 4.7SpaceXAI · statistically tied with #1

11 / 15

73.3%

48–89%

GPT-5.6 TerraOpenAI · statistically tied with #1

11 / 15

73.3%

48–89%

GPT-5.3-CodexOpenAI · statistically tied with #1

11 / 15

73.3%

48–89%

Muse Spark OpenMeta · statistically tied with #1

11 / 15

73.3%

48–89%

Kimi K3.1 ProMoonshotAI · statistically tied with #1

11 / 15

73.3%

48–89%

GPT-6 AstraOpenAI · statistically tied with #1

10 / 15

66.7%

42–85%

GPT-6.1 Terra ProOpenAI · statistically tied with #1

10 / 15

66.7%

42–85%

DeepSeek V4 Flash Vision ExpDeepSeek · statistically tied with #1

10 / 15

66.7%

42–85%

Gemini 3.5 Flash LiteGoogle · statistically tied with #1

10 / 15

66.7%

42–85%

Gemini 3.1 Pro PreviewGoogle · statistically tied with #1

10 / 15

66.7%

42–85%

GLM 5.3Z.ai · statistically tied with #1

10 / 15

66.7%

42–85%

GPT-6.1 SolOpenAI · statistically tied with #1

10 / 15

66.7%

42–85%

GPT-5.4OpenAI · statistically tied with #1

10 / 15

66.7%

42–85%

Claude Haiku 5.5Anthropic · statistically tied with #1

10 / 15

66.7%

42–85%

GPT-6.1 TerraOpenAI · statistically tied with #1

9 / 15

60.0%

36–80%

DeepSeek V4.1 Flash 1017DeepSeek · statistically tied with #1

9 / 15

60.0%

36–80%

GPT-6 LunaOpenAI · statistically tied with #1

9 / 15

60.0%

36–80%

Qwen3.8 Max (0902)Qwen · statistically tied with #1

9 / 15

60.0%

36–80%

o3OpenAI · statistically tied with #1

8 / 15

53.3%

30–75%

GLM 5.3 FlashZ.ai · statistically tied with #1

8 / 15

53.3%

30–75%

DeepSeek V4 Pro 0813DeepSeek · statistically tied with #1

8 / 15

53.3%

30–75%

GPT-6 SolOpenAI · statistically tied with #1

8 / 15

53.3%

30–75%

MiMo-V2.6-ProXiaomi · statistically tied with #1

8 / 15

53.3%

30–75%

Kimi K2.7 CodeMoonshotAI · statistically tied with #1

7 / 15

46.7%

25–70%

MiMo-V2.6-FlashXiaomi · statistically tied with #1

7 / 15

46.7%

25–70%

Llama 4 MaverickMeta

6 / 14

42.9%

21–67%

InklingThinking Machines

6 / 15

40.0%

20–64%

Step 5 PreviewStepFun

6 / 15

40.0%

20–64%

Mercury 2.5Inception

6 / 15

40.0%

20–64%

Phi 4Microsoft

6 / 15

40.0%

20–64%

Hy3Tencent

6 / 15

40.0%

20–64%

Qwen3 Coder NextQwen

6 / 15

40.0%

20–64%

Inkling SmallThinking Machines

6 / 15

40.0%

20–64%

Qwen3.8 27BQwen

6 / 15

40.0%

20–64%

Nemotron 3 UltraNVIDIA

6 / 15

40.0%

20–64%

MiniMax M3.1 FlashMiniMax

6 / 15

40.0%

20–64%

gpt-oss-120bOpenAI

6 / 15

40.0%

20–64%

Qwen3.8 2.4T A95BQwen

5 / 15

33.3%

15–58%

MiniMax M3MiniMax

5 / 15

33.3%

15–58%

Qwen3.7 PlusQwen

4 / 15

26.7%

11–52%

Nemotron 3.5 Super VLNVIDIA

4 / 15

26.7%

11–52%

gpt-oss-20bOpenAI

4 / 15

26.7%

11–52%

Ministral 3 8B 2512Mistral

4 / 15

26.7%

11–52%

Nemotron 3.5 LightningNVIDIA

4 / 15

26.7%

11–52%

Granite 4.2 8BIBM

4 / 15

26.7%

11–52%

Command ACohere

3 / 15

20.0%

7–45%

Mistral Large 4Mistral

3 / 15

20.0%

7–45%

Llama 4 ScoutMeta

2 / 15

13.3%

4–38%

Ministral 3 3B 2512Mistral

2 / 15

13.3%

4–38%

Solar Mini 4Upstage

2 / 15

13.3%

4–38%

Ministral 3 14B 2512Mistral

2 / 15

13.3%

4–38%

Llama 3.3 70B InstructMeta

2 / 15

13.3%

4–38%

Qwen3.6 35B A3BQwen

2 / 15

13.3%

4–38%

Ling 3.1 FlashinclusionAI

1 / 15

6.7%

1–30%

Granite 4.0 MicroIBM

1 / 15

6.7%

1–30%

Nova Micro 1.0Amazon

1 / 15

6.7%

1–30%

Solar Pro 4Upstage

1 / 15

6.7%

1–30%

LongCat 2.0Meituan

1 / 15

6.7%

1–30%

Mistral Medium 3.5Mistral

0 / 15

0.0%

0–20%

Reka Flash 3rekaai

0 / 15

0.0%

0–20%

Gemma 4 26B A4BGoogle

0 / 15

0.0%

0–20%

Gemma 4 31BGoogle

0 / 15

0.0%

0–20%

Qwen3.5 397B A17BQwen

0 / 15

0.0%

0–20%

Hermes 4 405BNous

0 / 15

0.0%

0–20%

Mistral Small 4Mistral

0 / 15

0.0%

0–20%

Trinity Large ThinkingArcee AI

0 / 15

0.0%

0–20%

Engineering: frequently asked questions

Which AI model is best at engineering?

Gemini 4 Argon answered the most engineering questions correctly: 13 of 14. With only 15 questions, 30 other models are statistically tied with the leader, so treat this ranking as a rough guide.

Where do the engineering questions come from?

14 from SuperGPQA (ODC-BY 1.0) and 1 from MMLU-Pro (MIT). Every question is multiple choice with ten options and one correct answer.

How hard are the engineering questions?

1 of the 15 questions were answered correctly by none of the 72 scored models, and 0 by all of them.

Best AI models for engineering

Engineering: score by model

How hard are these questions?

Engineering: frequently asked questions

Best AI models for engineering

Engineering: score by model

How hard are these questions?

Engineering: frequently asked questions