Look up Mistral's revenue and you get two numbers in two currencies.
Sacra estimated annualized revenue at $400 million in January 2026. Mistral itself forecast €1 billion in revenue for the full year.
The valuation is clearer. After its €3 billion Series D in September 2026, Mistral was valued at more than €21 billion.
Then there's Mistral Large 4, available as an API public preview since October 6, 2026.
Here are the key Mistral AI statistics for 2026 on funding and models, from Vibe to API pricing. The short version in three points:
- Mistral announced a €3B Series D in September 2026 at a post-money valuation above €21B.
- Sacra estimated annualized revenue at $400M in January 2026. Mistral forecast €1B in revenue for the full year.
- Large 4 has been in API public preview since October 6, 2026.
1. Mistral AI at a glance
Mistral's company profile says it was founded in April 2023. Its co-founders include Arthur Mensch, Guillaume Lample and Timothée Lacroix. In September 2026, the company announced a post-money valuation above €21B after its Series D.
2. Funding & valuation
Since its seed round in June 2023, Mistral's valuation has risen with every round. Most recently, it reached more than €21B.
2.1 Funding rounds since 2023
Mistral's disclosed rounds range from a €105M seed in June 2023 to a €3B Series D in September 2026. Reports on the €600M Series B said it combined equity and debt:
EUR
2.2 Post-money valuations by round
Reported post-money valuations also rose from round to round, with ASML leading the Series C. The bars show reported values in euros, the last bar as a lower bound. Series A is missing because its valuation was reported only in US dollars. TechCrunch cited Bloomberg at about $2B.
Mistral reported €3B in new capital for Series D at a post-money valuation above €21B. Samsung Electronics led the round, with the Scaleup Europe Fund and PSG Equity co-leading. Mistral called it the largest equity fundraising round completed by a European technology company.
3. Valuation in context
Mistral trails the large US labs by a wide margin. Among the European AI companies compared here, however, it leads.
3.1 Mistral next to Anthropic and OpenAI
The next chart compares the latest reported post-money round valuations for three private AI companies. For Mistral, it shows the September Series D (more than €21B, converted by TechCrunch on September 8 at that day's exchange rate):
3.2 European AI companies compared
Next to other European AI companies, Mistral's latest round is the largest of the three.
gradually.aiHelsing builds defense AI and Lovable an AI software-building product.
4. Revenue & growth
Sacra estimates annual recurring revenue at about $16M at the end of 2024, about $20M in January 2025, about $312M in December 2025 and $400M in January 2026:
Mistral forecast €1B in revenue for 2026. According to Le Monde on September 8, the forecast came from Mistral vice president Audrey Herblin-Stoop.
5. Vibe (formerly Le Chat) & usage
Mistral renamed Le Chat to Vibe on August 12, 2026, according to its help center. For the February 2025 mobile launch, Le Parisien, as cited by TechCrunch, reported 1 million app downloads in the first 14 days.
Mistral lists a free plan with limits on messages, web searches and coding sessions. For more figures on the AI industry, see my AI statistics.
Among entry-level and standard plans, Vibe Pro costs less than the $20 plans from ChatGPT, Claude, and Perplexity. Only ChatGPT Go is cheaper. The chart shows US list prices per month with monthly billing for individual users.
gradually.ai6. The model family
The following charts and the timeline select several important Mistral models. Large 4 has been available as an API public preview since October 6, and its weights appeared on Hugging Face on October 29, according to The Next Web.
6.1 Context windows of the models
Mistral advertises up to 1 million tokens of context for Large 4. Large 3 and Small 4 are described as 256K models:
6.2 Total and active parameters
The chart shows total parameters in billions. For mixture-of-experts models, the parameters active per token are considerably fewer.
The version-specific documentation lists 1.05T total parameters, 52B active parameters and 1 million tokens of context. The launch post gives a rounded 1 trillion total and the same 52 billion active parameters. OpenRouter lists 524,288 context tokens for its offering.
Small 4 also has differing parameter figures. Its model card lists 6.5B active, while the launch post says 6 billion, or 8 billion including input and output layers. Large 3 and Small 4 use Apache 2.0; Medium 3.5 uses Modified MIT. For a broader comparison, see my LLM statistics and open-source LLM directory.
6.3 The four models over time
Release, access and license of the four models in chronological order:
7. Benchmarks
Artificial Analysis published independent measurements for Large 4 Public Preview on October 6. It scored 50 points on the Cyber Index. On Intelligence Index v5, the October 27 leaderboard lists it at 25 points.
Mistral's launch post also gives 61.7% on DeepSWE v1.1 and explicitly attributes that score to the Artificial Analysis coding index. Le Monde reported 63% the same day.
The Arena leaderboard snapshot from October 2 placed Large 3 at 1,414 ± 3 Elo, rank 141 of 413 models, based on 79,452 votes.
On the same Artificial Analysis leaderboard, Large 4 Preview scores 25 points, at the bottom of this comparison. The top models from Anthropic, OpenAI, and Google reach 50 to 58 points. The index v5 leaderboard does not list Medium 3.5 and Small 4. The chart shows each model's value at its highest listed reasoning effort, retrieved on October 27, 2026.
gradually.ai8. API pricing
Mistral generally charges per million input and output tokens. The chart shows current list prices for input tokens; output prices appear in the labels:
Since October 29, 2026, Large 4 has cost $0.70 per million input tokens and $2.10 per million output tokens, with cached input at $0.07, according to Mistral's open-weights announcement. At launch on October 6, the list price was $1.36 input and $4.18 output. A 50% discount applied for the first two weeks, which meant $0.68 and $2.09.
Large 3 costs $0.50 and $1.50, Medium 3.5 $1.50 and $7.50, and Small 4 $0.15 and $0.60. See the API cost calculator for a broader cost calculation.
9. Context window & knowledge dates
Mistral advertises up to 1 million context tokens for Large 4, while OpenRouter lists 524,288 for its API offering. Large 3 and Small 4 are described as 256K models. Knowledge dates appear only in two published model system prompts.
The Small 3 system prompt says October 1, 2023, and the Small 4 prompt says November 1, 2024. The reviewed Large 4 documentation gives no comparable date.
10. Data centers & infrastructure
Mistral has announced several computing projects and capacity targets in France and elsewhere in Europe.
For Bruyères-le-Châtel in the Essonne department, Mistral secured around $830M in financing in March 2026 for 13,800 Nvidia GB300 GPUs. Eclairion operates the site, and Mistral announced 44 MW of power capacity.
Le Monde reported partial operation from January and expected full capacity later in the year. Its October 6 report said computing capacity was still coming online gradually.
The 200 MW target for the end of 2027 was set by Mistral in March. For Fouju, Bpifrance reported in June a secured first phase and a possible expansion to 200 MW. In August, Mistral set a longer-term target of up to 1 GW by 2030.
Mistral's May announcement also described a 10 MW inference site in Les Ulis planned for the third quarter of 2026. The hosting partner described the Paris South cluster.
gradually.aiThe bars show announced or secured capacity per site, with different statuses. For Fouju, Bpifrance reports a secured first phase of 96 MW.
Mistral's regional inference service lets customers choose regions such as Europe and the United States. Mistral notes that limited, protected subcontractors may be outside the selected region.
11. Open source & European AI sovereignty
Mistral publishes downloadable weights for part of its model range. Large 3 and Small 4 use Apache 2.0, while Medium 3.5 uses Modified MIT. Mistral released the Large 4 weights (about 1 trillion parameters, about 52 billion active) on Hugging Face, according to The Next Web on October 29, under the Mistral Ecosystem Open-Weight License, a modified MIT license. Anyone running a model-as-a-service business needs a separate commercial license.
France's Ministry of the Armed Forces announced a framework agreement in January 2026 that had been notified on December 16, 2025. It gives the armed forces, ministry services and supervised public bodies such as CEA, ONERA and SHOM access to Mistral solutions. The ministry's AI agency, AMIAD, manages the agreement.
Mistral has raised large funding rounds, put a new model into preview and announced several European computing projects. Whether Large 4 closes the gap in independent evaluations and every announced megawatt is online remains open.
