Llama 4 Scout

Llama 4 Scout is the lightest model in the next generation of the Llama 4 family, released on April 5, 2025. Llama 4 scout has a MoE architecture of 16 experts, 109 billion of which only 17 billion are permanently active. The model supports native multimodality. It is capable of simultaneously processing text and up to 5 images as input, allowing it to be used effectively in computer vision tasks such as object recognition, image description, and image-based question answering. Creative text generation, step-by-step reasoning, and programming are also among the model's strengths

That said, Llama scout can run efficiently on just a single H100 GPU, making it attractive for local deployment, and with more parameters, Llama scout outperforms lighter weight models in the segment on benchmarks.

But perhaps the most important feature of the model is its record-breaking context window support of up to 10 million tokens! which makes it ideal for working with large documents, code, multi-page queries and lengthy dialog sessions.

Announce Date: 05.04.2025
Parameters: 109B
Experts: 16
Activated at inference: 17B
Context: 10486K
Layers: 48, using full attention: 12
Attention Type: Chunked Attention
Developer: Meta AI
Transformers Version: 4.51.0
vLLM Version: 0.8.3
License: LLAMA 4

Use our pre-built public endpoints for free to test inference and explore Llama 4 Scout capabilities. You can obtain an API access token on the token management page after registration and verification.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.