Mistral Large 2 Environmental Impact

Energy Consumption

Mistral Large 2 consumes approximately 1.8 Wh per 400-token response, based on Mistral's official Life Cycle Assessment — one of the few AI companies to publish detailed environmental impact data. At 123B parameters, it is a dense model without MoE optimization.

Power Source & Carbon

Runs primarily on Azure infrastructure in France and EU regions. France's electricity grid has a carbon intensity of approximately 50 g CO2/kWh — among the lowest in Europe — due to its heavy reliance on nuclear power (~70%). This significantly reduces the carbon footprint per query compared to models running on US grids.

Water Usage

At approximately 45 mL per 400-token response, Mistral Large 2's water footprint is notably high according to Mistral's own LCA. This figure includes the full lifecycle water impact including upstream supply chain water use.

How much energy does Mistral Large 2 use per query?

Each Mistral Large 2 query consumes approximately 1.8 Wh of energy. This is 6x more than a traditional Google search (~0.3 Wh).

What is Mistral Large 2's carbon footprint?

Based on the carbon intensity of Azure France / EU regions, each query produces approximately 1.1 g of CO2. The grid in this region has a carbon intensity of 300 g CO2/kWh with 45% renewable energy.

How much water does Mistral Large 2 use?

Each query consumes approximately 45 mL of water, primarily used for cooling the data centers that process the request.

How does Mistral Large 2 compare to a Google search?

A Mistral Large 2 query uses 6x more than a Google search in terms of energy. A Google search uses approximately 0.3 Wh, while Mistral Large 2 uses 1.8 Wh.

Architecture

Dense Transformer (decoder-only)

Parameters

123B

Context window

128,000 tokens

Release date

2024-07-24

Open source

Yes