Memo sets GPT-5.6 Sol as Copilot's internal default; Microsoft cash generation fell 23% with Big Tech AI capex on track to top $700 billion this year.
Microsoft is telling developers on AI coding projects to default to OpenAI's GPT-5.6 Sol inside GitHub Copilot, an efficiency push that points its own engineers toward the models tied to its OpenAI investment over rival products.
Jay Parikh, executive vice president of Microsoft's CoreAI engineering group, spelled out the directive in a memo to employees this week that CNBC reviewed. "Internally, shifting more workloads to OpenAI models helps us get greater value from our token investment," Parikh wrote. Tokens are the units of AI processing; one token equals about three-quarters of a word. Technology news website 404 Media reported on the memo earlier.
Parikh, whose group runs GitHub and the Visual Studio tools, told staffers to default to OpenAI's flagship GPT-5.6 Sol when working in GitHub Copilot and to use that model most of the time. OpenAI released GPT-5.6 Sol in July. CoreAI will adjust the defaults as models and products change, Parikh wrote.
A Microsoft spokesperson confirmed the change by email. The company periodically updates "the default model settings in our internal tools to balance performance and efficient use of resources," the spokesperson said, and set GPT-5.6 Sol as the default for Microsoft's internal use of GitHub Copilot "while continuing to offer a range of model options." The default applies to Microsoft's own engineers; external Copilot customers keep the full model menu.
Microsoft has its own AI programming model and lets cloud customers choose from more than 11,000 models, including Anthropic's Claude. The directive still points staffers to OpenAI, where Microsoft holds intellectual-property rights from its early investment, extended through 2032 in a restructuring about nine months ago. Microsoft said in April it was ending revenue-sharing payments to OpenAI.
The memo arrives after a stretch of so-called tokenmaxxing, when developers were encouraged to run up large token bills without much regard for what they produced. Open-weight models, largely out of China, have gained ground because they are cheaper to access than frontier models and can be tuned and hosted on infrastructure of the user's choosing.
Wall Street is starting to demand more from the biggest AI spending commitments. Capital expenditures from Microsoft, Amazon, Alphabet and Meta are expected to top $700 billion combined this year. Free cash flow across the group fell in the latest quarter and turned negative for Amazon and Alphabet. Microsoft's cash generation dropped 23% from a year earlier, a milder decline than its peers.
Microsoft's stock rose 22% last week after earnings, its best weekly showing since 1999. Shares are up about 1% for the year, trailing most megacap peers.
AlphaScala's risk model scores MSFT 72 out of 100, a Moderate label; the stock traded at $489.80, down 0.61% on the day. Wells Fargo's Michael Turrin said Microsoft stands to benefit in AI infrastructure and applications.
Microsoft's internal preference for OpenAI models sits alongside a deepening commercial relationship with Anthropic. Microsoft agreed late last year to invest up to $5 billion in Anthropic and released Copilot Cowork, a product that runs on Anthropic's Claude models. Anthropic committed to spend $30 billion on Microsoft Azure cloud services. The companies have not made public assurances on intellectual property.
CEO Satya Nadella has said companies can lower costs by separating AI services from models. Anthropic's Claude Code, a software development agent, offers only Anthropic models. GitHub Copilot lists models from Anthropic, Google, OpenAI, xAI and Microsoft itself.
Microsoft moved early to have AI models compose code, and GitHub Copilot now counts 50 million users. Newer products such as Cursor have taken share in the fast-moving coding-assistant market.
Parikh said Microsoft divisions manage their own token budgets, and CoreAI has not yet set budgets for individual teams or employees. He asked each employee to find one example of AI spending that helped a customer or business outcome, and one that did not.
"If you have a big idea or big project that will need significant token usage, have a quick chat with your manager," Parikh wrote.
Drafted by a large language model from the source reporting linked above, then screened by automated publishing checks. It is not read by a journalist before publication. Some articles cite our Alpha Score. Verify prices and figures against the original source. Educational coverage, not personalized advice.