GPT-4

OpenAI

GPT-4 is a powerful multimodal model that accepts both image and text inputs, producing sophisticated, human-quality text as output. Its performance on a range of professional and academic tests is comparable to that of a human.

Model Specifications

Technical details and capabilities of GPT-4

Core Specifications

32.8K / 32.8K

Input / Output tokens

December 30, 2022

Knowledge cutoff date

June 12, 2023

Release date

Capabilities & License

Multimodal Support

Supported

Web Hydrated

Yes

License

Proprietary

Resources

Research Paper

https://arxiv.org/abs/2303.08774

API Reference

https://platform.openai.com/docs/api-reference/chat

Playground

https://platform.openai.com/playground

Code Repository

https://github.com/openai/gpt-4

Performance Insights

Check out how GPT-4 handles various AI tasks through comprehensive benchmark results.

100

96.3

AI2 Reasoning Challenge (ARC)

96.3

(96%)

95.3

HellaSwag

95.3

(95%)

Uniform Bar Exam

(90%)

SAT Math

(89%)

LSAT

(88%)

87.5

Winogrande

87.5

(88%)

86.4

MMLU

86.4

(86%)

80.9

DROP

80.9

(81%)

74.5

MGSM

74.5

(75%)

HumanEval

(67%)

MATH

(42%)

35.7

GPQA

35.7

(36%)

AI2 Reasoning Challenge (ARC)

HellaSwag

Uniform Bar Exam

SAT Math

LSAT

Winogrande

MMLU

DROP

MGSM

HumanEval

MATH

GPQA

Model Comparison

See how GPT-4 stacks up against other leading models across key performance metrics.

86.4

MMLU - GPT-4

86.4

(96%)

MMLU - Claude 3 Sonnet

(88%)

75.2

MMLU - Claude 3 Haiku

75.2

(84%)

78.9

MMLU - Phi-3.5-MoE-instruct

78.9

(88%)

69.8

MMLU - GPT-3.5 Turbo

69.8

(78%)

MMLU - Phi-3.5-mini-instruct

(77%)

HumanEval - GPT-4

(74%)

HumanEval - Claude 3 Sonnet

(81%)

75.9

HumanEval - Claude 3 Haiku

75.9

(84%)

70.7

HumanEval - Phi-3.5-MoE-instruct

70.7

(79%)

HumanEval - GPT-3.5 Turbo

(76%)

62.8

HumanEval - Phi-3.5-mini-instruct

62.8

(70%)

35.7

GPQA - GPT-4

35.7

(40%)

40.4

GPQA - Claude 3 Sonnet

40.4

(45%)

33.3

GPQA - Claude 3 Haiku

33.3

(37%)

36.8

GPQA - Phi-3.5-MoE-instruct

36.8

(41%)

30.8

GPQA - GPT-3.5 Turbo

30.8

(34%)

30.4

GPQA - Phi-3.5-mini-instruct

30.4

(34%)

MATH - GPT-4

(47%)

43.1

MATH - Claude 3 Sonnet

43.1

(48%)

38.9

MATH - Claude 3 Haiku

38.9

(43%)

59.5

MATH - Phi-3.5-MoE-instruct

59.5

(66%)

43.1

MATH - GPT-3.5 Turbo

43.1

(48%)

48.5

MATH - Phi-3.5-mini-instruct

48.5

(54%)

74.5

MGSM - GPT-4

74.5

(83%)

83.5

MGSM - Claude 3 Sonnet

83.5

(93%)

75.1

MGSM - Claude 3 Haiku

75.1

(83%)

58.7

MGSM - Phi-3.5-MoE-instruct

58.7

(65%)

56.3

MGSM - GPT-3.5 Turbo

56.3

(63%)

47.9

MGSM - Phi-3.5-mini-instruct

47.9

(53%)

MMLU

HumanEval

GPQA

MATH

MGSM

GPT-4

Claude 3 Sonnet

Claude 3 Haiku

Phi-3.5-MoE-instruct

GPT-3.5 Turbo

Phi-3.5-mini-instruct

Detailed Benchmarks

Dive deeper into GPT-4's performance across specific task categories. Expand each section to see detailed metrics and comparisons.

Coding

HumanEval

Claude 3.5 Sonnet

93.7%

Llama 3.1 8B Instruct

72.6%

72.0%

70.7%

68.0%

67.0%

Phi-3.5-mini-instruct

62.8%

Gemma 2 27B

51.8%

Gemma 2 9B

40.2%

Ministral 8B Instruct

34.8%

Current model

Other models

Avg (63.4%)

Reasoning

HellaSwag

95.4%

95.3%

89.0%

88.6%

87.6%

86.4%

85.9%

85.2%

Current model

Other models

Avg (89.2%)

DROP

DeepSeek-R1

92.2%

Llama 3.1 405B Instruct

84.8%

83.4%

83.1%

83.1%

80.9%

80.2%

79.7%

Llama 3.1 70B Instruct

79.6%

Llama 3.1 8B Instruct

59.5%

Current model

Other models

Avg (80.7%)

Knowledge

MMLU

91.8%

Llama 3.1 405B Instruct

87.3%

86.9%

86.8%

86.5%

86.4%

86.2%

Llama 3.2 90B Instruct

86.0%

Llama 3.3 70B Instruct

86.0%

Llama 3.2 3B Instruct

63.4%

Current model

Other models

Avg (84.7%)

GPQA

87.7%

36.9%

36.8%

36.4%

35.9%

35.7%

33.3%

Llama 3.2 11B Instruct

32.8%

Llama 3.2 3B Instruct

32.8%

Qwen2 7B Instruct

25.3%

Current model

Other models

Avg (39.4%)

MATH

o3-mini

97.9%

Qwen2.5-Coder 7B Instruct

46.6%

43.1%

43.1%

42.3%

42.0%

38.9%

36.6%

32.6%

Current model

Other models

Avg (47.0%)

Non categorized

Winogrande

GPT-4

87.5%

Command R+

85.4%

Qwen2 72B Instruct

85.1%

Llama 3.1 Nemotron 70B Instruct

84.5%

Qwen2.5 32B Instruct

82.0%

Gemma 2 9B

80.6%

Mistral NeMo Instruct

76.8%

Ministral 8B Instruct

75.3%

Current model

Other models

Avg (82.2%)

MGSM

92.0%

83.5%

82.6%

80.6%

75.1%

74.5%

Llama 3.2 11B Instruct

68.9%

Phi-3.5-MoE-instruct

58.7%

Llama 3.2 3B Instruct

58.2%

Phi-3.5-mini-instruct

47.9%

Current model

Other models

Avg (72.2%)

Providers Pricing Coming Soon

We're working on gathering comprehensive pricing data from all major providers for GPT-4. Compare costs across platforms to find the best pricing for your use case.

OpenAI

Anthropic

Google

Mistral AI

Cohere

Share your feedback

Hi, I'm Charlie Palars, the founder of Deepranking.ai. I'm always looking for ways to improve the site and make it more useful for you. You can write me through this form or directly through X at @palarsio.

Your feedback helps us improve our service

GPT-4

Model Specifications

Core Specifications

Capabilities & License

Resources

Performance Insights

Model Comparison

Detailed Benchmarks

Coding

HumanEval

Reasoning

HellaSwag

DROP

Knowledge

MMLU

GPQA

MATH

Non categorized

Winogrande

MGSM

Providers Pricing Coming Soon

Share your feedback

Stay Ahead with AI Updates