Mistral Large 4: what an open European model means for SMEs
In brief
According to Next, on 29 October 2026 Mistral announced general availability of the final Large 4 model and published its weights as open source. The company says extra weeks of reinforcement learning raised scores, including 41% on Terminal-Bench 4.0 among open-weight models. The article describes a multimodal model of about 1,000 billion parameters, with a focus on cybersecurity and pricing.
Source : Next - « [MàJ] Mistral lance la version finale de Large 4 et publie ses poids ouverts »
What it changes for an SME
For an SME of 10 to 250 people, a French open-weight model trained, according to Mistral, in its own European data centres may matter if you want an EU-based option. Hosting a roughly trillion-parameter model yourself is usually impractical. Prices quoted in the original coverage (USD 1.36 per million input tokens and USD 4.18 per million output tokens) should be checked for the final release, and Artificial Analysis described the model as very verbose, which can raise the bill. Nothing in this news creates a new compliance duty by itself; it is a vendor choice on cost, where inference runs, and how you govern cybersecurity-related use.
What you can do
- Run a small cost test on your real prompts before switching, because verbosity can change the invoice.
- Ask where inference is hosted, not only where the model was trained.
- If you trial cybersecurity features, restrict access and keep an audit trail.
Let's talk about the time you could save.
In 20 minutes, we find where AI could save you time or money. Free, no commitment.