Microsoft has unveiled a family of seven in-house AI models at its annual Build developer conference in San Francisco, in the clearest sign yet that the tech giant is moving to reduce its dependence on the AI companies it has poured billions into.
Mustafa Suleyman, Microsoft AI chief executive, said that after tuning its models for consulting firm McKinsey, the company outperformed OpenAI's GPT-5.5 on quality — with what it projects as ten times better cost efficiency, based on public pricing data scaled across model sizes.
"We believe the time has come for every company to move from consuming a frontier model to fully participating at the frontier," said chief executive Satya Nadella at the conference.
The headline release is MAI-Thinking-1, Microsoft's first reasoning model, trained from scratch on clean, commercially licensed data without distillation from third-party systems.
A mid-sized model with 35 billion active parameters and a 256,000-token context window, it is designed for complex multi-step instructions, long-context reasoning and code generation.
Alongside it, the company launched MAI-Code-1-Flash, a coding model that converts text descriptions into source code for applications and websites, now rolling out across GitHub Copilot and Visual Studio Code.
By running its own models on Azure infrastructure, Microsoft can sidestep the fees it currently pays to third-party providers — and pass the savings to developers.
According to Microsoft, in blind evaluations run by Surge, its independent human rating partner, MAI-Thinking-1 was preferred over Anthropic's Claude Sonnet 4.6, and the company says it matches Claude Opus 4.6 on coding benchmarks.
Quantum leap
TL;DR: Qubits, the fundamental unit of quantum computing, are notoriously fragile: even minor temperature shifts or vibrations can knock them off course.
Also at Build, Microsoft announced that its Majorana 2 quantum chip is 1,000 times more reliable than its predecessor — a milestone the company said brings it within striking distance of a commercially useful quantum computer.
Qubits, the fundamental unit of quantum computing, are notoriously fragile: even minor temperature shifts or vibrations can knock them off course. The Majorana 2 chip addresses this directly.
Qubits on the new chip survive for an average of 20 seconds, compared to milliseconds on the original, an improvement the company likened to upgrading from a phone that needs daily charging to one that lasts several years.
"We will have a quantum machine in 2029 that can solve commercially viable, reasonable problems," said Zulfi Alam, corporate vice president of Microsoft Quantum. The chip currently has just 12 qubits, while a useful machine would require millions.
Microsoft's approach centres on so-called topological qubits, based on the properties of a quasi-particle first theorised in the 1930s by Italian physicist Ettore Majorana.
The path has not been smooth — the company was forced to retract a 2018 paper in the journal Nature claiming evidence for the particle — but it has pressed on, and the second-generation chip improves on the first partly by replacing aluminium with lead as a superconductor.
The chip and its supporting research have not yet been peer reviewed, and some physicists have called for more information.
IPO race heats up
TL;DR: Microsoft's push for model independence comes as the companies it has committed billions to prepare for blockbuster stock market debuts.
Microsoft's push for model independence comes as the companies it has committed billions to prepare for blockbuster stock market debuts.
Anthropic, the AI lab behind Claude, filed confidentially for an initial public offering on 1 June, just days after raising $65 billion (€59bn) in a Series H funding round that pushed its valuation to $965 billion (€877bn). OpenAI is also readying its own confidential filing.
Microsoft has committed $13 billion (€11.8bn) to OpenAI and invested up to $5 billion (€4.5bn) in Anthropic, while making both companies' models available through Azure.
Curated by Sofia Andersson
