Llama 3.2 3B Instruct

Meta Llama

Llama 3.2 3B Instruct is a cutting-edge large language model excelling in on-device applications. With a 128K token context window, it's ideal for edge computing tasks like summarization, instruction following, and rewriting, delivering state-of-the-art performance directly on your device.

Model Specifications

Technical details and capabilities of Llama 3.2 3B Instruct

Core Specifications

3.2B Parameters

Model size and complexity

9000.0B Training Tokens

Amount of data used in training

128.0K / 128.0K

Input / Output tokens

November 30, 2023

Knowledge cutoff date

September 24, 2024

Release date

Capabilities & License

Multimodal Support

Not Supported

Web Hydrated

No

License

Llama 3.2 Community License

Resources

API Reference

https://github.com/meta-llama/llama-models

Playground

https://llama.meta.com/llama-downloads

Code Repository

https://github.com/meta-llama/llama-models

Performance Insights

Check out how Llama 3.2 3B Instruct handles various AI tasks through comprehensive benchmark results.

90

68

45

23

0

84.7

NIH/Multi-needle

84.7

(94%)

78.6

ARC-C

78.6

(87%)

77.7

GSM8K

77.7

(86%)

77.4

IFEval

77.4

(86%)

69.8

HellaSwag

69.8

(78%)

67

BFCL V2

67

(74%)

63.4

MMLU

63.4

(70%)

63.3

InfiniteBench/En.MC

63.3

(70%)

58.2

MGSM

58.2

(65%)

48

MATH

48

(53%)

40.1

Open-rewrite

40.1

(45%)

34.3

Nexus

34.3

(38%)

32.8

GPQA

32.8

(36%)

19.8

InfiniteBench/En.QA

19.8

(22%)

19

TLDR9+ (test)

19

(21%)

NIH/Multi-needle

ARC-C

GSM8K

IFEval

HellaSwag

BFCL V2

MMLU

InfiniteBench/En.MC

MGSM

MATH

Open-rewrite

Nexus

GPQA

InfiniteBench/En.QA

TLDR9+ (test)

Detailed Benchmarks

Dive deeper into Llama 3.2 3B Instruct's performance across specific task categories. Expand each section to see detailed metrics and comparisons.

Math

GSM8K

97.1%

Gemini 1.5 Flash

86.2%

Phi-3.5-mini-instruct

86.2%

Qwen2.5-Coder 7B Instruct

83.9%

Qwen2 7B Instruct

82.3%

Llama 3.2 3B Instruct

77.7%

74.0%

68.6%

Current model

Other models

Avg (82.0%)

Reasoning

HellaSwag

Qwen2.5 32B Instruct

85.2%

Phi-3.5-MoE-instruct

83.8%

Mistral NeMo Instruct

83.5%

Qwen2.5-Coder 32B Instruct

83.0%

81.9%

Qwen2.5-Coder 7B Instruct

76.8%

Llama 3.2 3B Instruct

69.8%

Phi-3.5-mini-instruct

69.4%

Current model

Other models

Avg (79.2%)

Knowledge

MMLU

69.7%

Llama 3.1 8B Instruct

69.4%

69.2%

Phi-3.5-mini-instruct

69.0%

Mistral NeMo Instruct

68.0%

Qwen2.5-Coder 7B Instruct

67.6%

Ministral 8B Instruct

65.0%

Llama 3.2 3B Instruct

63.4%

Current model

Other models

Avg (67.7%)

MATH

97.9%

50.6%

49.7%

Qwen2 7B Instruct

49.6%

Phi-3.5-mini-instruct

48.5%

Llama 3.2 3B Instruct

48.0%

Qwen2.5-Coder 7B Instruct

46.6%

Claude 3 Sonnet

43.1%

43.1%

32.6%

Current model

Other models

Avg (51.0%)

GPQA

87.7%

35.9%

35.7%

33.3%

Llama 3.2 11B Instruct

32.8%

Llama 3.2 3B Instruct

32.8%

32.3%

30.8%

Llama 3.1 8B Instruct

30.4%

Qwen2 7B Instruct

25.3%

Current model

Other models

Avg (37.7%)

Non categorized

IFEval

Claude 3.7 Sonnet

93.2%

83.9%

83.3%

Mistral Small 3

82.9%

Llama 3.1 8B Instruct

80.4%

Llama 3.2 3B Instruct

77.4%

Qwen2.5 7B Instruct

71.2%

63.0%

61.3%

Current model

Other models

Avg (77.4%)

ARC-C

94.8%

92.4%

90.2%

Llama 3.2 3B Instruct

78.6%

Qwen2.5 32B Instruct

70.4%

Qwen2.5 14B Instruct

67.3%

Current model

Other models

Avg (82.3%)

Nexus

Llama 3.1 405B Instruct

58.7%

Llama 3.1 70B Instruct

56.7%

Llama 3.1 8B Instruct

38.5%

Llama 3.2 3B Instruct

34.3%

Current model

Other models

Avg (47.0%)

MGSM

92.0%

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%

56.3%

Phi-3.5-mini-instruct

47.9%

Current model

Other models

Avg (66.5%)

Providers Pricing Coming Soon

We're working on gathering comprehensive pricing data from all major providers for Llama 3.2 3B 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