Microsoft is cutting rival AI models out of its low-code builder. From December 2, prompts built in Copilot Studio, Power Automate and Power Apps will no longer run on Anthropic’s Claude or xAI’s Grok, after Microsoft told administrators it is retiring both vendors’ models from its prompt builder. Existing prompts will switch automatically to OpenAI’s newer GPT-6 models, with no change to tier pricing. Makers now have just under four weeks to test the replacements before the change takes effect.
## The models leaving, and what replaces them
The swap runs across all three pricing tiers. At Basic, GPT-4.1 mini gives way to GPT-6 Luna. At Standard, GPT-4.1, GPT-5 chat and GPT-5.3 chat go, with Claude Sonnet 4.6 and Grok 4.1 Fast reasoning, all replaced by ChatGPT-latest. At Premium, GPT-5 reasoning, GPT-5.2 reasoning and Claude Opus 4.6 give way to GPT-6 Reasoning.
Microsoft says the migration is automatic and makers do not need to pick a replacement to receive it. But the company warns that answers can change after the move, including wording, formatting and how instructions are followed, so it is asking teams to review and test prompts before December 2 and to fix workflows that depend on a specific output.
The change covers Power Platform prompts only. It does not touch the model picker in GitHub Copilot, which runs its own retirement schedule, or the Copilot chat apps, where model choice works differently. The distinction matters because the same model names appear in several Microsoft products, and a retirement in one place does not mean the model is gone everywhere.
## Why It Matters
Model choice was one of Copilot Studio’s selling points. Letting makers pick Claude for careful reasoning or Grok for fast responses meant Microsoft’s builder could serve teams that had already standardised on a rival model. Removing that choice narrows the builder to a single vendor family and makes Microsoft’s OpenAI partnership the only route for AI inside Power Platform prompts.
The deadline is also tight for the teams most affected. Prompts sit inside production apps, approval flows and customer-facing agents, so a model that answers differently can break a workflow quietly. Microsoft’s own advice to test first is the right call, but four weeks is a short window for organisations with dozens of prompts to check.
For administrators, the change is also a governance story. A prompt that quietly switches models can alter how customer data is handled or how strictly instructions are followed, and Microsoft’s notice puts the testing burden on makers rather than offering a longer transition. With no phased rollout, teams that skip the checks will discover any differences in production.
## Our Read
This is consolidation, not housekeeping. Microsoft is pruning older GPT versions at the same time, which is routine, but dropping Claude and Grok is a statement about who controls the AI inside its business platform. The timing fits a wider pattern: across its products, Microsoft has been steering users toward the GPT-6 family while rival models quietly lose their seats.
The losers here are not Anthropic or xAI, whose models stay available through their own APIs and through [GitHub Copilot’s model routing](https://geroundtable.com/github-copilot-fusion-model-routing/). The cost falls on makers who chose Claude or Grok deliberately and now face re-testing, and on Microsoft’s pitch that Copilot Studio is the neutral place to build agents.
If the GPT-6 replacements perform as promised, most users will barely notice. The real test is the hardest prompts: Claude Opus 4.6 was the Premium tier’s reasoning option, and if GPT-6 Reasoning answers those prompts differently, the teams that picked Opus will feel it first.
## Background
Copilot Studio lets business users build AI agents and add AI prompts to apps and flows without writing code, and the prompt builder has long offered a choice of models across its tiers. Microsoft has been tightening model line-ups across Copilot this autumn, including [opening its Copilot harness to outside builders](https://geroundtable.com/microsoft-copilot-harness-foundry-builders/), as it pushes GPT-6 across its products.
Source: [Microsoft’s Message Center notice](https://mc.merill.net/message/MC1536101) · [publication record](https://mc.changecockpit.cloud/change/MC1536101)
0 Comments