What is the carbon footprint of Mistral's Le Chat, and what did Mistral's own lifecycle study find?
For sustainability teams and analysts who need a number for the Mistral models their company uses. Sources checked on 20 November 2026. Factor version soma-ai-ef-2026.11-dataset-v3.1.0. Factors last updated . Location-based grid averages, not renewable contracts. Estimates from published benchmarks.
Page updated
Part of the SOMA AI Emission Factors cluster.
What did Mistral's own lifecycle study find for Le Chat?
Mistral AI, working with Carbone 4 and the French agency ADEME, reports 1.14 g CO₂e and 45 mL of water for one 400-token Le Chat response. The study is a lifecycle assessment of Mistral Large 2 that follows the GHG Protocol Product Standard and ISO 14040/44, published on 22 July 2025 (Mistral AI). The figure is the marginal impact of inference, including upstream impacts such as server manufacturing, and it excludes users' terminals. Mistral reports training separately, at 20.4 kt CO₂e. We found no per-query figure from Anthropic, OpenAI or Meta, which makes Mistral and Google the two providers with something to check against.
How do Mistral models compare in different regions?
Mistral's study gives one figure for one response length, so it does not say what a smaller or larger model costs, or where the request ran. SOMA AI (somaai.earth) fills that gap with estimates from the open SOMA AI Emission Factors (SAEF). SAEF assigns each Mistral model a class from its published size or tier, and assumes eu-west-3 as the default region. These are estimates, not Mistral measurements.
| Model | SAEF class | Wh | eu-west-3 (EU West - France) | eu-north-1 (EU North - Sweden) | us-east-1 (US East - N. Virginia) |
|---|---|---|---|---|---|
| Ministral 8B | Class A | 0.03 | 0.0029 g | 0.0025 g | 0.01 g |
| Mistral Small | Class B | 0.099 | 0.011 g | 0.0094 g | 0.036 g |
| Mistral Large | Class C | 0.37 | 0.043 g | 0.037 g | 0.13 g |
Mistral Large is 15 times Ministral 8B in the same region, and the same Large exchange is 3.1 times higher in US East - N. Virginia than in EU West - France. The table uses an exchange of 928 input and 282 output tokens, not Mistral's 400-token response.
Why is Mistral's figure so different from the SAEF estimate?
For Mistral Large, Mistral's 1.14 g is about 27 times our estimate of 0.043 g in EU West - France. The response lengths differ, and the boundaries differ too. SAEF also counts hardware manufacturing, data-centre construction and amortised training, on a location-based grid. The Mistral summary does not name the model behind the figure or break it down, so we cannot say where the difference comes from. The general case for why published figures disagree is in the published-figures guide.
How does Mistral compare with ChatGPT, Claude and Gemini?
SAEF gives models of the same class and the same region the same number, so it cannot show whether one brand is lower than another. Same class and same region give the same estimate. Mistral and Google publish figures on their own boundaries, and the two are not comparable with each other. Sam Altman's 0.34 Wh for the average ChatGPT query, from his personal blog with no stated boundary, is not an OpenAI figure.
| SAEF class | Claude | OpenAI | Google Gemini | Mistral |
|---|---|---|---|---|
| Class A | claude-haiku-4, 0.01 g | gpt-5-mini, 0.01 g | gemini-2.5-flash, 0.01 g | ministral-8b, 0.01 g |
| Class B | claude-sonnet-4, 0.036 g | gpt-4o, 0.036 g | gemini-2.5-pro, 0.036 g | mistral-small, 0.036 g |
| Class C | claude-opus-4, 0.13 g | gpt-5, 0.13 g | No Gemini model in this class | mistral-large, 0.13 g |
More models are on the model comparison, and the Claude and Gemini pages are here and here.
What should a company report for Mistral?
Take token counts per model from Mistral's usage export or from your cloud platform, and use the region in the export where there is one. Report the class, the region, the factor version and the range, and say that Mistral's own figure covers one 400-token Le Chat response on a marginal-inference boundary. SOMA AI estimates this use and does not measure it. The method is in the Scope 3 guide and on the methodology page, and the free calculator gives a first estimate without an account.
Short answers
- What did Mistral's own lifecycle study find for Le Chat?
- Mistral AI, with Carbone 4 and ADEME, reports 1.14 g CO2e and 45 mL of water for one 400-token Le Chat response, counting marginal inference impacts and excluding users' terminals. It reports training of Mistral Large 2 separately, at 20.4 kt CO₂e.
- How does SOMA AI's estimate for Mistral Large compare with Mistral's figure?
- SOMA AI estimates, from its open SAEF emission factors, 0.043 g CO2e for one Mistral Large exchange of 1,210 tokens in the EU West - France region. Mistral's 1.14 g is about 27 times that. The two use different response lengths and boundaries, and the Mistral summary does not name the model behind its figure, so we cannot say where the difference comes from.
- Is Mistral lower or higher than ChatGPT, Claude and Gemini?
- SAEF cannot tell them apart inside a size class, because it assigns each model a class from its provider's tier or size. Mistral is one of only two providers in our comparison with a per-query figure we could check, and that figure is on Mistral's own boundary, so it does not rank Mistral against the others.
- Where does SOMA AI assume Mistral serves from?
- SAEF assumes eu-west-3, EU West - France, for Mistral models. That is an assumption, not a statement of where Mistral runs a request. If you call a model through a cloud platform, use the region in the usage export.
Keep reading
- Which AI model has the lowest carbon footprint? GPT, Claude, Gemini, Llama, DeepSeek and Mistral compared. Provider figures beside SAEF estimates
- What is the carbon footprint of Claude, and how does it compare with ChatGPT? Haiku, Sonnet and Opus by region
- What is the carbon footprint of Gemini, and how does it compare with ChatGPT and Claude? Google's figure, Flash and Pro
- What is the carbon footprint of one ChatGPT prompt? Published per-prompt figures and their basis
