Mistral: open-weight frontier models, built in Europe
Conventional wisdom held that frontier models would only come from a handful of US labs. A small team in Paris shipping open-weight models looked like a long shot.
The Bet
Mistral AI was founded in Paris in 2023. The consensus at the time was that building competitive large language models required enormous budgets, and that the frontier would belong to a few American companies.
Moonshot’s bet was that a small, exceptionally strong research team could reach the frontier with far fewer resources, and that releasing open-weight models would build trust and adoption faster than keeping everything closed.
Openness was not a charitable choice. It was a distribution strategy.
Moonshot’s View
We believed many companies and governments would want models they could inspect, run and adapt themselves. Openness was not a charitable choice; it was a distribution strategy.
We also believed geography was a weaker moat than people assumed. World-class talent exists in many places, and the best teams do not need to be in one zip code to do frontier work.
Room for Disagreement
Critics argued that open-weight models are hard to monetise, that smaller labs cannot keep up with the compute budgets of the largest players, and that the gap to the frontier would only widen.
Each point has weight. Our disagreement was about efficiency: we thought a focused team could do more with less, and that being open would earn a place in the stack that money alone cannot buy.
Notable
Mistral, in brief.
- Sector
- Open models
- Founded
- 2023
- Founders
- Arthur Mensch, Guillaume Lample and Timothée Lacroix
- Moonshot backed
- Early
- The contrarian call
- Frontier AI could be open, efficient and built in Europe
- Money raised
- $6.61B to date
- Company value
- $24B
- Moonshot 25
- No. 24 of 25 overall
Source: money raised and company value from Caplight, as of Nov 17, 2026. View source
Pitch Moonshot
Building something the crowd doesn’t believe in yet?
Tell us in about two minutes. No warm intro needed, and every pitch is read by the team.