# Llama 3

The open-source AI models you can fine-tune, distill and deploy anywhere. Choose from our collection of models: Llama 3.1, Llama 3.2, Llama 3.3.

[Download models](/content/llama-downloads/index.html)

## Llama 3 models

Llama includes multilingual text-only models (1B, 3B), including quantized versions, text-image models (11B, 90B) and Llama 3.3 70B model offering similar performance to the Llama 3.1 405B model, allowing developers to achieve greater quality and performance on text-based applications at a fraction of the cost.

### MOST RECENT

#### Llama 3.3

[View documentation](/content/docs/model-cards-and-prompt-formats/llama3_3/index.html)

Llama 3.3 is a text-only 70B instruction-tuned model that provides enhanced performance relative to Llama 3.1 70B–and relative to Llama 3.2 90B when used for text-only applications. Moreover, for some applications, Llama 3.3 70B approaches the performance of Llama 3.1 405B.

### VARIATIONS

70B Leading performance at a fraction of the cost.

[Download](/content/llama-downloads/index.html)

### OLDER MODELS

#### Llama 3.2

[View documentation](/content/docs/model-cards-and-prompt-formats/llama3_2/index.html)

Llama 3.2 11B and 90B Vision-enabled models plus lightweight 1B and 3B models optimized for on-device and edge deployment.

### VARIATIONS

1B & 3B Light-weight, efficient models you can run everywhere.

[Download](/content/llama-downloads/index.html)  
11B & 90B Multimodal and can reason on high resolution images.

[Download](/content/llama-downloads/index.html)

#### Llama 3.1

[View documentation](/content/docs/model-cards-and-prompt-formats/llama3_1/index.html)

The open source AI model you can fine-tune, distill and deploy anywhere. Our instruction-tuned model is available in 8B, 70B and 405B versions.

### VARIATIONS

8B Light-weight, ultra-fast model you can run anywhere.

[Download](/content/llama-downloads/index.html)  
405B Flagship foundation model driving widest variety of use cases.

[Download](/content/llama-downloads/index.html)

## Do more with Llama

On-device

Use our 1B or 3B models for on device applications such as summarizing a discussion from your phone or calling on-device tools like calendar.

[Learn more](/content/docs/model-cards-and-prompt-formats/llama3_2/index.html)

Multimodal

Use our 11B or 90B models for image use cases such as transforming an existing image into something new or getting more information from an image of your surroundings.

[Learn more](/content/docs/model-cards-and-prompt-formats/llama3_1/index.html)

Llama Stack

Seamlessly build agentic applications from a comprehensive toolchain.

[Learn more](https://github.com/meta-llama/llama-stack/blob/main/docs/getting_started.ipynb)

[View all capabilities](/content/docs/overview/index.html)

## Start building with Llama 3

[Download models](/content/llama-downloads/index.html) [View documentation](/content/docs/overview/index.html)

## SAFETY

### Protections in the era of generative AI.

Comprehensive system-level protections proactively identify and mitigate potential risks, empowering developers to more easily deploy generative AI responsibly.

[Protection tools accessible to everyone \n\nLearn more](/content/llama-protections/index.html)

[Enabling AI Defenders \n\nLearn more](/content/llama-protections/ai-defenders/index.html)

## Llama 3 tuned benchmarks

Llama 3.3 instruction

Lightweight instruction

Vision instruction

| Category           | Benchmark                          | Pricing*                     |
|--------------------|-----------------------------------|-----------------------------|
| General            | MMLU Chat (0-shot, CoT)          | $0.1 (1M Input tokens)      |
|                    | MMLU PRO (5-shot, CoT)           | $0.4 (1M Output tokens)     |
| Instruction Following| IFEval                           |                             |
| Code               | HumanEval (0-shot)               |                             |
|                    | MBPP EvalPlus (base) (0-shot)    |                             |
| Math               | MATH (0-shot, CoT)               |                             |
| Reasoning          | GPQA Diamond (0-shot, CoT)       |                             |
| Tool use           | BFCL v2 (0-shot)                 |                             |
| Long context       | NIH/Multi-needle                  |                             |
| Multilingual       | Multilingual MGSM (0-shot)       |                             |

| Model              | MMLU Chat | MMLU PRO | IFEval | HumanEval | MBPP EvalPlus | MATH | GPQA Diamond | BFCL v2 | NIH/Multi-needle | Multilingual MGSM | Pricing |
|--------------------|----------|----------|--------|------------|----------------|------|---------------|---------|-------------------|------------------|--------|
| Llama 3.1 70B     | 86.0     | 66.4     | 87.5   | 80.5       | 86.0           | 67.8 | 48.0          | 77.5    | 97.5              | 86.9             | $0.1  |
|                    |          |          |        |            |                |      |               |         |                   |                  | $0.4  |
| Llama 3.3 70B     | 86.0     | 68.9     | 92.1   | 88.4       | 87.6           | 77.0 | 50.5          | 77.3    | 97.5              | 91.1             | $0.1  |
|                    |          |          |        |            |                |      |               |         |                   |                  | $0.4  |
| Amazon Nova Pro    | 85.9     | -        | 92.1   | 89.0       | -              | 76.6 | -             | -       | -                 | -                | $0.80 |
|                    |          |          |        |            |                |      |               |         |                   |                  | $3.20 |
| Llama 3.1 405B    | 88.6     | 73.4     | 88.6   | 89.0       | 88.6           | 73.9 | 49.0          | 81.1    | 98.1              | 91.6             | $1.0  |
|                    |          |          |        |            |                |      |               |         |                   |                  | $1.8  |
| Gemini Pro 1.5    | 87.1     | 76.1     | 81.9   | 89.0       | 87.8           | 82.9 | 53.5          | 80.3    | 94.7              | 89.6             | $1.30 |
|                    |          |          |        |            |                |      |               |         |                   |                  | $5.0  |
| GPT-4o            | 87.5     | 73.8     | 84.6   | 86.0       | 83.9           | 76.9 | 47.5          | 74.0    | -                 | 90.6             | $2.5  |
|                    |          |          |        |            |                |      |               |         |                   |                  | $10.0 |
| Claude 3.5       | 88.9     | 77.8     | 89.3   | 93.7       | 86.8           | 78.3 | 65.0          | 79.3    | 99.4              | 92.8             | $3.0  |
|                    |          |          |        |            |                |      |               |         |                   |                  | $15.0 |

\* API Pricing based on publicly available data on [Artificial Analysis](https://artificialanalysis.ai/) as of 12/3/24.