# Llama 4

Leading Intelligence. Unrivaled speed and efficiency. The most accessible and scalable generation of Llama is here.

## Llama 4 models

Our models are optimized for easy deployment, cost efficiency, and performance that scales to billions of users.

[View documentation](/content/docs/model-cards-and-prompt-formats/llama4/index.html)

#### Llama 4 Scout

Class-leading natively multimodal model that offers superior text and visual intelligence, single H100 GPU efficiency, and a 10M context window for seamless long document analysis. [Download](/content/llama-downloads/?preselect_models=LLAMA_4_16E_OMNI_V2&&utm_source=llama-home-scout&utm_medium=llama-referral&utm_campaign=llama-utm&utm_offering=llama-scout&utm_product=llama/index.html)

#### Llama 4 Maverick

Industry-leading natively multimodal model for image and text understanding with groundbreaking intelligence and fast responses at a low cost. [Download](/content/llama-downloads/?preselect_models=LLAMA_4_128E_OMNI_V2&&utm_source=llama-home-maverick&utm_medium=llama-referral&utm_campaign=llama-utm&utm_offering=llama-maverick&utm_product=llama/index.html)

## Llama 4 capabilities

- **Natively multimodal**
- **Unparalleled long context**
- **Expert image grounding**

All Llama 4 models are designed with native multimodality, leveraging early fusion that allows us to pre-train the model with large amounts of unlabeled text and vision tokens - a step change in intelligence from separate, frozen multimodal weights.

### How to guides:
[Video capabilities](/content/docs/how-to-guides/vision-capabilities/index.html)

### Video tutorials:
[Build a Multimodal WhatsApp Chatbot](/content/resources/videos/build-multi-modal-whatsapp-chatbot-with-llama-4/index.html)

## Start building with Llama 4
[View documentaion](/content/docs/model-cards-and-prompt-formats/llama4/index.html) [Github](https://github.com/meta-llama/llama-cookbook/blob/main/getting-started/build_with_llama_4.ipynb)

### 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](/content/llama-protections/index.html)

[Llama Defenders Program](/content/llama-protections/ai-defenders/index.html)

## Llama 4 benchmark

| Task                     | Metric                          | Llama 4 Maverick | Llama 4 Scout |
|--------------------------|---------------------------------|------------------|----------------|
| Image Reasoning          | MMMU                            | 73.4             | 69.4           |
| MathVista                |                                 | 73.7             | 70.7           |
| Image Understanding       | ChartQA                        | 90               | 88.8           |
| DocVQA (test)           |                                 | 94.4             | 94.4           |
| Coding                   | LiveCodeBench  (10.01.2024 - 02.01.2025) | 43.4             | 32.8           |
| Reasoning & Knowledge    | MMLU Pro                       | 80.5             | 74.3           |
| DocVQA (test)           |                                 | 69.8             | 57.2           |
| Multilingual             | Multilingual MMLU              | 84.6             | -              |
| Long Context             | MTOB (half book)              | eng->kgv/kgv->eng | 54.0 / 46.4   |
|                          | eng->kgv/kgv->eng               | 42.2 / 36.6     |                |
|                          | MTOB (full book)              | eng->kgv/kgv->eng | 50.8 / 46.7   |
|                          | eng->kgv/kgv->eng              | 39.7 / 36.3     |                |

## Methodology & Notes
1. For Llama model results, we report 0 shot evaluation with temperature = 0 and no majority voting or parallel test time compute. For high-variance benchmarks (GPQA Diamond, LiveCodeBench), we average over multiple generations to reduce uncertainty.
2. Specialized long context evals are not traditionally reported for generalist models, so we share internal runs to showcase llama's frontier performance.
3. $0.19/Mtok (3:1 blended) is our cost estimate for Llama 4 Maverick assuming distributed inference. On a single host, we project the model can be served at $0.30 - $0.49/Mtok (3:1 blended).