On 6 October, two launches came out that you can use straight away as a Dutch professional. Mistral released a large European model whose weights will be released soon. Anthropic put Claude in a sidebar in Google Docs, Sheets and Slides. Both make AI easier to use, but they work out very differently for your GDPR assessment.

Mistral Large 4: large, European, soon on your own servers

Mistral has released Mistral Large 4 as a public preview. The model has 1 trillion parameters, of which 49 billion are active per task (a so-called mixture of experts). It processes text and images and, according to Mistral, supports more than 160 languages. Via Mistral Studio and the API you pay $1.36 per million input tokens and $4.18 per million output tokens.

Two points are important for your privacy assessment:

  • Mistral trained the model on its own GPUs in European data centres. According to the announcement, the European service runs from start to finish without other digital service providers and under European law.
  • According to Mistral, the weights will be released at the end of October. You can then run the model in your own environment, so your data never leaves your organisation. The announcement does not yet say under which licence this will happen. Check this before you deploy it commercially.

Note: Mistral says it is still testing the model with security partners and governments. The benchmark figures in the announcement come from Mistral itself. Once independent scores are available, you will see them on /modellen/ and on /mistral/.

For anyone working with sensitive data, such as lawyers, GPs, banks or the government, a large model you can host yourself is a serious option. Do bear in mind that you will need substantial hardware: a model of this size cannot run on an ordinary server.

Claude as a sidebar in Docs, Sheets and Slides

Anthropic has released Claude for Google Workspace as a public beta for all paid Claude subscriptions. You install the add-on from the Google Workspace Marketplace. Afterwards, Claude can rewrite text and correct spelling in Docs, write formulas and build tables in Sheets, and create new slides on your existing layout in Slides. With the "ask before editing" setting, you see every change before it is applied. You can also paste a link to a Google file into a Claude chat and edit the file from within Claude.

Handy, because you no longer need to copy and paste. But it also means the contents of your documents go directly to Anthropic. What Anthropic does with that data depends on your subscription. Pro and Max fall under the consumer terms: there, chats are used for training by default unless you turn that off. Team and Enterprise do not work that way. You can read how it works per subscription at /claude/ and /veiligheid/.

If you work in an organisation with Google Workspace, first agree who may install the add-on and which documents it may access. You should not simply let an add-on that falls under a personal Pro subscription read a spreadsheet containing client or patient data.

In brief

  • Mistral Large 4 is now available as a preview via the API. The weights arrive at the end of October; the licence is not yet known.
  • Claude now works in Google Docs, Sheets and Slides for all paid subscriptions, in beta.
  • If you use Claude in Workspace with work documents, choose Team or Enterprise, or turn off training on Pro and Max.

Every Friday: what changed in AI for your field

New models, prices, oversight and tools, with sources and what it means for you. One email a week, unsubscribe with one click.

View a sample

We only use your address for this newsletter. You confirm first by email.