Microsoft deploys new AI models in race against OpenAI
The company unveiled models for coding and cybersecurity, with an emphasis on lower costs.
Listen to this article
There is no audio version yet. Request one and an AI voice will read the article aloud.
Read by an AI voice.
Microsoft has announced new AI models for coding, cybersecurity and other specialised tasks. With them, the company is seeking to reduce its dependence on OpenAI and compete directly with OpenAI and Anthropic.
The announcement came on Tuesday during Microsoft Ignite in San Francisco. Microsoft presented MAI Code 2 and MAI Code 2 Flash, among others—models intended to write software and support developers. According to Microsoft, MAI Code 2 achieves about 98 per cent of the coding quality of GPT-6.1 Sol at roughly 40 per cent of the price.
Microsoft also released models for cybersecurity. The company says they can help detect and analyse digital attacks. This is a performance claim made by Microsoft itself, not an independently established result.
The move fits a broader shift within the AI market. Large software companies are not only building applications on top of other companies’ models, but are also developing their own models to gain better control over costs, availability and product integration. At the same time, Microsoft continues to offer OpenAI models through its Azure AI Foundry cloud platform.
For business users, the price per use may become more important than an overall ranking. With frequent use—for example, for code review, documentation and testing software—costs can mount. A cheaper model may therefore be attractive, even if it does not deliver the best result for every task.
The announcement says little so far about performance in day-to-day development environments. The test results reported by Microsoft do not necessarily show how the models perform when used by programmers over extended periods. For users, the programming language, human oversight and the sensitivity of the source code may also matter.
Microsoft presents MAI Code 2 as a model for programming tasks. Dutch companies considering the model can weigh up security, data processing, costs and its suitability for their tasks.
Also show the region and a close-up
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The announcement, model names and performance figures cited by Microsoft can be found in the company’s publication and model card. The text explicitly attributes the comparisons and cost claims to Microsoft.
- confirmed Microsoft announced MAI Code 2 and other AI models on 17 November. — Microsoft AI published the announcement and model card. source
- confirmed Microsoft says MAI Code 2 achieves about 98 per cent of GPT-6.1 Sol at roughly 40 per cent of the price. — These figures appear in Microsoft’s announcement. source
- confirmed Reuters reported that Microsoft presented models for coding and cybersecurity. — Reuters report via MarketScreener. source
Editor's note
The model performance and price comparisons are company claims; independent comparisons in practice are not yet available.Sources
- Seven new world class models at scalable price-performance — Microsoft AI
- Microsoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAI — Reuters via MarketScreener
- MAI Code 2 and MAI Code 2 Flash Model Card — Microsoft AI