Microsoft is racing to make OpenAI optional
The New Stack Alex Wilhelm ● Covered by 75 sources
Microsoft is quietly swapping OpenAI's models for its own inside Excel and GitHub Copilot. It's not about beating GPT-5, it's about not paying OpenAI for every query.
Based on reporting by The New Stack, Alex Wilhelm — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Satya Nadella has been busy on Twitter lately, and it's not just vibes — Microsoft is shipping homegrown AI models into its own products at a pace that suggests real urgency. Last week Redmond laid out two data points that matter more than they first appear.
First, an internal model now running inside Excel is, according to Microsoft, on par with GPT-5.6 for the most common spreadsheet tasks, while costing less to run. Take the parity claim with a grain of salt — Microsoft doesn't need to out-benchmark the frontier, it needs something good enough that the average Excel user never notices the difference. The cost angle is the real story. Every query routed to a Microsoft-built model instead of GPT-5.6 is money Microsoft doesn't have to send to OpenAI or Anthropic, two companies it happens to be simultaneously investing in, competing with, and buying compute from.
Second, and more concretely, there's MAI-Code-1-Flash, a model Microsoft post-trained inside the GitHub Copilot harness. Millions of developers are apparently already using it day to day, and Microsoft says it beats GPT-5.4 Mini and Claude Haiku 4.5 in VS Code by roughly 10 percentage points on code-accept rate, while burning fewer tokens and holding onto users slightly better.
None of this flips the AI leaderboard. Microsoft isn't claiming frontier supremacy, and it shouldn't — that's not the game it's playing. What it's demonstrating is that it can build production-grade models tuned to its own products using its own mountain of enterprise coding and workplace data, data no outside lab gets to touch because Microsoft is the lab now. Redmond calls this feedback loop its 'hill-climbing machine,' and if it compounds the way Microsoft hopes, OpenAI starts looking less like a dependency and more like an option.
My take — AI-written commentary, not fact-checked reporting
This is the most rational move in the entire AI stack right now: stop paying frontier-lab prices for tasks that don't need frontier intelligence. Every hyperscaler eventually builds its own 'good enough' model to cut the tax it pays upstream — AWS did it with Titan, Google did it years ago with its own stack, and Microsoft was always going to follow once its OpenAI bill got embarrassing. The interesting tell here isn't the benchmark numbers, it's that Microsoft is willing to publish them at all, which tells you how much leverage it thinks it's building against its favorite frenemy.
Read more about this at: The New Stack