LLaMA 4 Scout Environmental Impact

Energy Consumption

LLaMA 4 Scout activates only 17B of its 109B total parameters per token via MoE routing across 16 experts. It fits on a single H100 GPU with int4 quantisation, making it remarkably efficient for its capability level. Estimated at ~0.6 Wh per short query — similar to a small dense model despite its large total parameter count.

Power Source & Carbon

As an open-source model that fits on a single GPU, Scout can be self-hosted on diverse infrastructure. The carbon impact depends entirely on the deployment location. Meta's own data centres run about 60% on renewable energy.

Water Usage

At approximately 1.1 mL per query, Scout's water footprint is low. When self-hosted on personal hardware, water consumption drops to effectively zero.

How much energy does LLaMA 4 Scout use per query?

Each LLaMA 4 Scout query consumes approximately 0.60 Wh of energy. This is 2x more than a traditional Google search (~0.3 Wh).

What is LLaMA 4 Scout's carbon footprint?

Based on the carbon intensity of Self-hosted (varies), each query produces approximately 0.24 g of CO2. The grid in this region has a carbon intensity of 475 g CO2/kWh with 27% renewable energy.

How much water does LLaMA 4 Scout use?

Each query consumes approximately 1 mL of water, primarily used for cooling the data centers that process the request.

How does LLaMA 4 Scout compare to a Google search?

A LLaMA 4 Scout query uses 2x more than a Google search in terms of energy. A Google search uses approximately 0.3 Wh, while LLaMA 4 Scout uses 0.60 Wh.

Architecture

Multimodal Transformer (MoE, 16 experts, iRoPE)

Parameters

109B

Context window

10,000,000 tokens

Release date

2025-04-05

Open source

Yes

Training data cutoff

2025-02