Mistral Large 4 in preview: 1 trillion parameters, weights to follow at the end of October
On 6 October, Mistral announced Mistral Large 4, a multimodal model with 1 trillion parameters, of which 49 billion are active per task. You can now use it as a preview via Mistral Studio and the API, at $1.36 per million input tokens and $4.18 per million output tokens. Mistral trained the model in its own European data centres and, according to the announcement, will release the weights at the end of this month.
What does this mean? Once the weights are available, you can run a large European model on your own servers, so sensitive data does not go to a US provider.
Claude as a sidebar in Google Docs, Sheets and Slides (public beta)
Anthropic has released Claude for Google Workspace as a public beta for the paid Pro, Max, Team and Enterprise plans. The add-on reads and edits only the file that is open: in Sheets it writes formulas and creates pivot tables and charts, and in Slides it builds slides on your existing layout. By default, Claude asks for your approval before making any change, and administrators can roll out or block the add-on via the Google Admin console.
What does this mean? If your organisation works in Google Workspace, you can use Claude without copying and pasting. Your administrator does need to decide first whether the add-on is permitted.
Berlin privacy regulator warns about personalised ChatGPT ads
The Berlin data protection authority (BlnBDI) points to considerable privacy risks if OpenAI uses your chats to personalise ads. This only happens if you give consent, and it concerns the free ChatGPT and ChatGPT Go. If you do not give consent, you receive general ads which, according to the regulator, still use the content of your current chat.
What does this mean? If you or your staff use free ChatGPT or Go for work, do not consent to personalised ads. In addition, keep sensitive topics in a temporary chat or use a subscription without ads.
Google launches image model Nano Banana 2.1, half the price per image
On 6 October, Google released Nano Banana 2.1 in the Gemini app, AI Mode in Search, Google Ads and developer tools such as Google AI Studio. Via the API, a 1K image costs $0.0336, half the price of Nano Banana 2 ($0.067), and a 4K image costs $0.0756. Google reports better editing of cut-outs and more consistent people and objects across multiple edits.
What does this mean? If you produce a lot of imagery for campaigns or product pages, generating via the API becomes cheaper. Do the maths for heavy editing workflows, though, as the other rates have changed too.
