How much does a Mistral prompt consume?
One Mistral response (Le Chat assistant, about 400 tokens) accounts for 1.14 g of CO₂e and 45 mL of water, according to the life-cycle assessment Mistral AI published with France's environment agency ADEME on July 22, 2025. It is the first full LCA released by an AI vendor: it counts the whole life cycle, not just data-center electricity, and Mistral publishes no electricity (Wh) figure.
Figures published on July 22, 2025, per ~400-token response (full life-cycle assessment) · Sources: Mistral AI : analyse de cycle de vie de Le Chat, avec l'ADEME (juillet 2025)
Why 45 mL of water, when Google reports 0.26 mL for Gemini?
Because the two figures don't measure the same thing. Google counts the water evaporated on site to cool its data centers while answering. Mistral also counts the water behind upstream electricity generation, server manufacturing and the rest of the life cycle. Both values can be right at the same time: the scope changes, not necessarily the facts. That's exactly why every figure on this site states what it includes.
What exactly does Mistral's assessment count?
The study, run with the consultancy Carbone 4 and ADEME, then peer-reviewed by Resilio and Hubblo, follows ISO 14040/44, the GHG Protocol and AFNOR's frugal-AI methodology. It also quantifies training: after 18 months, training and running Mistral Large 2 accounted for 20.4 kilotonnes of CO₂e and 281,000 m³ of water. Brought down to one 400-token response, the marginal impact is 1.14 g of CO₂e and 45 mL of water.
Does a Mistral prompt consume more than a ChatGPT prompt?
They can't be compared directly: Mistral publishes a life-cycle assessment (broad scope), while OpenAI and Epoch AI publish operational inference figures (~0.3 Wh, i.e. ~0.32 mL of direct water and ~0.2 g of CO₂ on an average grid). On an equivalent scope, models of comparable size sit in the same order of magnitude. The real lesson of Mistral's LCA: a model 10 times bigger generates impacts roughly 10 times larger for the same output.
And in electricity (Wh), how much does Mistral use?
Mistral has not published a watt-hour value: its study reports CO₂e, water and resource depletion (antimony). For scale, figures published for comparable models range from 0.24 Wh (Google, median Gemini prompt) to 0.34 Wh (OpenAI) per text prompt; our converter files standard prompts at 0.3 Wh.
A Mistral prompt in everyday actions
The figures published for Mistral, converted with the site's reference actions (every action value is sourced, see the Sources page):
- 1 email with attachment = 17 Mistral prompts
- 1 km by car = 149 Mistral prompts
- 1 glass of water = 5.6 Mistral prompts
- One 5-minute shower = 1,111 Mistral prompts