Another day, another AI model. At least that’s how it seems. Just between Anthropic’s Claude, OpenAI’s ChatGPT, and Microsoft’s own MAI models, there’s a constant flow of new and updated AI models for Microsoft 365 apps.
Be aware of changes in AI models because the new software can behave in a different way. Sometimes better, occasionally worse, but often just differently. The same prompt and sources can get a very different response with even a small change in the AI model.
Claude Opus 5
Like earlier Opus models, v5 comes in four levels from Haiku (fast, quick), Sonnet, Opus, and the highest level Fable. Only Sonnet (best for most tasks) and Opus (for more complex jobs) are available in the Microsoft 365 apps.
In Excel 365, choose from Sonnet v5 or Opus v5 with Opus 4.8 available for compatibility.
In Word 365 and PowerPoint 365 only Opus 4.8 or Sonnet v5 are available.
Unlike ChatGPT, the Copilot labels are the same as the ones used by Anthropic. That makes it easier to judge which model to use.
How to get Claude v5 in Microsoft 365 apps
If you have Copilot with your Microsoft 365 plan then new AI models should appear automatically.
If not, there’s a few things to try. Restart the app, each Microsoft 365 app checks online for the latest model information when the app starts. Failing that, make sure the Microsoft 365 apps are up to date because changes in Copilot might not appear unless the software is up to date.
Which to choose?
Of the Claude options, Sonnet is the best choice for simple tasks like a small rewrite, writing or fixing a formula.
Switch to Opus for more complex tasks like proofing or analyzing a whole document or spreadsheet.
Most people will stick with one model, most likely the more powerful one (e.g. Opus v5). That’s easier than switching models.
Office Watch avoids “Auto” because you don’t know which model will be used and Microsoft is likely to choose the cheapest model over the one best suited to the task.
Opus v5 is good for in-depth research, analysis, document production, or multi-step business tasks rather than pure coding, this is the largest real gain in the release. By some measures, Opus v5 is almost on par with Claude’s most complex and expensive model, Fable.
However early testers found some annoyances, which is typical for a new model and a reason why both Anthropic and Microsoft offer the preceding model as a backup.
