Llama 3.2 90B Instruct

Meta Llama

Llama 3.2 90B is a powerful multimodal language model adept at understanding visual information, interpreting images, and generating descriptive captions. With a substantial context window of 128,000 tokens, it excels in both image comprehension and creative generation. Optimized for use on edge and mobile devices, this model delivers cutting-edge performance in a variety of image-related tasks.

Model Specifications

Technical details and capabilities of Llama 3.2 90B Instruct

Core Specifications

90.0B Parameters

Model size and complexity

128.0K / 128.0K

Input / Output tokens

November 30, 2023

Knowledge cutoff date

September 24, 2024

Release date

Capabilities & License

Multimodal Support

Supported

Web Hydrated

No

License

llama3.2

Resources

API Reference

https://huggingface.co/meta-llama/Llama-3.2-90B-Vision-Instruct

Code Repository

https://huggingface.co/meta-llama/Llama-3.2-90B-Vision-Instruct

Performance Insights

Check out how Llama 3.2 90B Instruct handles various AI tasks through comprehensive benchmark results.

100

75

50

25

0

92.3

AI2 Diagram

92.3

(92%)

90.1

DocVQA

90.1

(90%)

86.9

MGSM

86.9

(87%)

86

MMLU

86

(86%)

85.5

ChartQA

85.5

(86%)

78.1

VQAv2

78.1

(78%)

73.5

Text VQA

73.5

(74%)

68

MATH

68

(68%)

60.3

MMMU

60.3

(60%)

57.3

MathVista

57.3

(57%)

56.8

InfographicsQA

56.8

(57%)

46.7

GPQA

46.7

(47%)

45.2

MMMU-Pro

45.2

(45%)

AI2 Diagram

DocVQA

MGSM

MMLU

ChartQA

VQAv2

Text VQA

MATH

MMMU

MathVista

InfographicsQA

GPQA

MMMU-Pro

Detailed Benchmarks

Dive deeper into Llama 3.2 90B Instruct's performance across specific task categories. Expand each section to see detailed metrics and comparisons.

Knowledge

MMLU

91.8%

86.8%

86.5%

86.4%

86.2%

Llama 3.2 90B Instruct

86.0%

Llama 3.3 70B Instruct

86.0%

85.9%

85.9%

Llama 3.2 3B Instruct

63.4%

Current model

Other models

Avg (84.5%)

MATH

97.9%

70.2%

Claude 3.5 Haiku

69.4%

69.3%

Mistral Small 3.1 24B

69.3%

Llama 3.2 90B Instruct

68.0%

61.6%

60.1%

Qwen2 72B Instruct

59.7%

32.6%

Current model

Other models

Avg (65.8%)

GPQA

87.7%

Qwen2.5 32B Instruct

49.5%

Qwen2.5 72B Instruct

49.0%

48.0%

46.9%

Llama 3.2 90B Instruct

46.7%

Qwen2.5 14B Instruct

45.5%

Mistral Small 3

45.3%

Qwen2 72B Instruct

42.4%

Qwen2 7B Instruct

25.3%

Current model

Other models

Avg (48.6%)

Non categorized

VQAv2

80.9%

78.6%

Llama 3.2 90B Instruct

78.1%

Current model

Other models

Avg (79.2%)

Text VQA

81.5%

80.2%

Llama 3.2 90B Instruct

73.5%

Current model

Other models

Avg (78.4%)

DocVQA

Claude 3.5 Sonnet

95.2%

93.3%

93.2%

92.8%

90.7%

Llama 3.2 90B Instruct

90.1%

Llama 3.2 11B Instruct

88.4%

85.6%

85.6%

Current model

Other models

Avg (90.5%)

MMMU

Gemini Pro 2.5 Experimental

81.7%

63.2%

Mistral Small 3.1 24B

62.8%

Gemini 1.5 Flash

62.3%

61.7%

Llama 3.2 90B Instruct

60.3%

59.4%

56.2%

Gemini 1.5 Flash 8B

53.7%

0.0%

Current model

Other models

Avg (56.1%)

ChartQA

Claude 3.5 Sonnet

90.8%

88.1%

Mistral Small 3.1 24B

86.2%

85.7%

Llama 3.2 90B Instruct

85.5%

Llama 3.2 11B Instruct

83.4%

Phi-3.5-vision-instruct

81.8%

81.8%

76.1%

Current model

Other models

Avg (84.4%)

AI2 Diagram

Llama 3.2 90B Instruct

92.3%

Llama 3.2 11B Instruct

91.1%

Current model

Other models

Avg (91.7%)

MMMU-Pro

Mistral Small 3.1 24B

49.3%

Qwen2-VL-72B-Instruct

46.2%

Llama 3.2 90B Instruct

45.2%

Llama 3.2 11B Instruct

33.0%

Current model

Other models

Avg (43.4%)

MathVista

74.9%

Claude 3.5 Sonnet

67.7%

Gemini 1.5 Flash

65.8%

63.8%

63.8%

Llama 3.2 90B Instruct

57.3%

56.7%

Gemini 1.5 Flash 8B

54.7%

52.8%

0.0%

Current model

Other models

Avg (55.7%)

MGSM

92.0%

89.3%

88.5%

87.5%

87.0%

Llama 3.2 90B Instruct

86.9%

Claude 3.5 Haiku

85.6%

Claude 3 Sonnet

83.5%

Gemini 1.5 Flash

82.6%

Phi-3.5-mini-instruct

47.9%

Current model

Other models

Avg (83.1%)

Providers Pricing Coming Soon

We're working on gathering comprehensive pricing data from all major providers for Llama 3.2 90B Instruct. 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

Stay Ahead with AI Updates

Get insights on Gemini Pro 2.5, Sonnet 3.7 and more top AI models