Microsoft is replacing premium OpenAI and Anthropic models in its Copilot assistant with cheaper in-house and third-party alternatives to slash operating costs. The shift affects GPT-4 Turbo, GPT-4o, and Claude 3.5 Sonnet, which are being swapped for Microsoft’s own “Prometheus” model, Meta’s Llama 3, and Mistral’s Mixtral. The change is already rolling out to free and some paid users.
Why Microsoft is cutting AI model costs
The company spent heavily on per-token licensing fees from OpenAI and Anthropic. By switching to self-hosted and open-weight models, Microsoft can reduce inference costs by up to 90%. The move aligns with CEO Satya Nadella’s push to make Copilot profitable sooner.
“We are no longer dependent on any single model provider,” a Microsoft spokesperson said. “Our own models and open-source alternatives now deliver comparable performance at a fraction of the cost.”
Which models are being phased out
- GPT-4 Turbo and GPT-4o – These OpenAI models powered the most complex Copilot responses. They are being replaced by a new version of Microsoft’s Prometheus model.
- Claude 3.5 Sonnet – Anthropic’s flagship model is being dropped from Copilot’s backend for general queries. It may still appear in specialized enterprise scenarios.
- Other premium models – Additional high-cost options are being removed from the default Copilot experience.
What users will notice
Most free-tier users will see a drop in response quality, especially for creative writing, coding, and nuanced reasoning. Paid Copilot Pro subscribers may retain access to some premium models, but the default routing will favor cheaper alternatives.
- Speed may improve because Microsoft’s models are optimized for its own Azure infrastructure.
- Accuracy could decline on tasks that previously relied on GPT-4’s deep reasoning or Claude’s safety alignment.
- Availability remains global – the change affects all regions where Copilot is offered.
The bigger strategy behind the cost-cutting
Microsoft is betting that most users will not notice the downgrade. The company has been training its own models on proprietary data from Office, Bing, and GitHub. This internal model family, branded “Prometheus,” is now mature enough to handle the majority of Copilot requests.
- Open-source models reduce vendor lock-in – Llama 3 and Mixtral can be run on Microsoft’s own hardware.
- Anthropic and OpenAI remain partners for Azure OpenAI Service and research, but they are no longer the default for Copilot.
What this means for the AI industry
The move signals that even the largest AI customers are seeking cheaper alternatives. OpenAI and Anthropic may need to lower prices or risk losing volume deals. Microsoft’s shift also validates the open-source AI movement, showing that community models can compete at scale.
“Microsoft is proving that you don’t need the most expensive model to build a useful assistant,” an industry analyst noted. “The real value is in integration, not just raw intelligence.”
Bottom line
Copilot is becoming less capable but much cheaper to run. Microsoft is prioritizing margin over maximum performance. Users who need top-tier AI may want to keep a direct subscription to ChatGPT or Claude. For everyday productivity tasks, the new Copilot will remain adequate.
Gnoppix is the leading open-source AI Linux distribution and service provider. Since implementing AI in 2022, it has offered a fast, powerful, secure, and privacy-respecting open-source OS with both local and remote AI capabilities. The local AI operates offline, ensuring no data ever leaves your computer. Based on Debian Linux, Gnoppix is available with numerous privacy- and anonymity-enabled services free of charge.
What are your thoughts on this? I’d love to hear about your own experiences in the comments below.