Microsoft steps up competition with OpenAI and Anthropic: its own AI models and agents

On July 29, 2026, Microsoft presented Wall Street with its own AI models, agent harnesses and even a competitor to the Mythos product. It is a signal that the company is competing ever more openly with its partners OpenAI and Anthropic.

What Microsoft showed

Microsoft rolled out three classes of in-house AI work at once: language models, harnesses (frameworks for running AI agents) and a direct competitor to the Mythos product. TechCrunch reported this on July 29, 2026, following the company's conversations with investors.

• Announcement date: July 29, 2026
• Microsoft's own language AI models
• Agent harnesses: frameworks for running and orchestrating agents
• A competitor to a product called Mythos
• Audience: Wall Street, with a message of continued growth

Microsoft framed this as part of a strategy for further growth, not a one-off experiment. Betting on its own stack, from foundation models to agent infrastructure, means the company wants to own the entire chain, not just the storefront in the form of Copilot.

The presentation came amid conversations with investors about business results. Microsoft tied its in-house AI work to a promise of continued growth and made clear that it sees AI as its own source of revenue, not just a line item of spending on partners.

What is an agent harness?

An agent harness is the software layer that turns a language model into a working agent. It gives the model tools, memory and access to a browser and code, and it manages the "think, act, verify" loop. On July 29, 2026, Microsoft showed investors its own such framework for the first time, not just models.

Control over this layer is critical for enterprise scenarios, from Copilot to workflow automation. Whoever owns the harness decides which models and tools get plugged in under the hood, and this is exactly where Microsoft wants to stop depending on other people's solutions. The fight is not over a single model but over the entire layer on top of it.

Why this matters for OpenAI and Anthropic

Microsoft remains a key partner of both labs while becoming an increasingly explicit competitor. For years the company integrated other firms' models into its products and invested in OpenAI. Now it is publicly showing that it is building its own alternative, down to a competitor to one specific product, Mythos.

For OpenAI and Anthropic, this means the largest distributor of their technology in the enterprise segment is turning into a direct rival. The more Microsoft relies on its own models and harnesses, the less it depends on its partners, and the fiercer the fight for the same corporate AI budgets.

Copilot and Microsoft 365 enterprise subscriptions are the main channel through which businesses get access to cutting-edge AI. If the models under the hood of that channel turn out to be Microsoft's own rather than OpenAI's, the balance of power in the enterprise AI market could shift faster than analysts expected.

"Microsoft presented its own AI models, harnesses and even a

Mythos competitor," says TechCrunch's July 29, 2026 report.

What this means

Microsoft's shift from "partner-distributor" to independent player is accelerating. On July 29, 2026, the company showed Wall Street its bet on a proprietary AI stack more openly than ever before. For the market, it is a signal that the line between allies and competitors in AI is blurring for good, and that enterprise customers will soon be choosing not between "Microsoft and OpenAI" but between the Microsoft stack and everyone else. The coming quarters will show how quickly Microsoft replaces other companies' models with its own inside its own products.