Ministral 3

Ministral 3 is Mistral AI's small-model line, released in the v25.12 wave alongside Mistral Large 3 and published in three sizes: 14B, 8B and 3B parameters. All three carry Apache 2.0 weights, which is the whole point of the line — these are models you are meant to download, fine-tune and run on your own hardware, not only call through an API. Mistral describes them as offering best-in-class text and vision capabilities for their size, with the 3B billed as the tiny end of the range. The pricing on Mistral's own API shows how the line is positioned: $0.20 per million tokens for 14B, $0.15 for 8B and $0.10 for 3B — and unusually, input and output cost the same in each case, which is what you would expect from a model priced by compute rather than by perceived value of the answer. For comparison, Mistral Small 4 costs $0.15 in and $0.60 out, and Mistral Medium 3.5 costs $1.50 and $7.50. The practical role is the one that rarely makes headlines: on-device and on-premises work, classification and extraction at volume, drafting inside applications where a round trip to a frontier model is too slow or too expensive, and any deployment where the data cannot leave the building. Mistral does not publish context window figures for these models on its public overview page, so this profile does not state them. The catalogue treats the three sizes as one entry because they were released together as one family under one version tag; if Mistral splits them or gives one of them a separate life, they will get their own profiles.

▸Videos

No videos yet.

▸Reviews