Microsoft Replaces OpenAI and Anthropic to Cut Costs

Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Microsoft Replaces OpenAI and Anthropic to Cut Costs
Microsoft is replacing the AI models from OpenAI and Anthropic with its own internal models across several Copilot products, including Excel and Outlook.
Microsoft's internal models, called MAI, are already handling tens of thousands of queries per week in Excel and Outlook, according to Bloomberg. These applications previously relied more heavily on the models from OpenAI and Anthropic.
Although the internal models currently manage only a small fraction of total queries, Microsoft aims to continue reducing its expenses related to third-party AI over time. The MAI models are also available in GitHub Copilot, and a proprietary transcription model is expected to launch in Teams soon.
At the Build conference, Microsoft unveiled seven new AI models, including MAI-Thinking 1, its first reasoning model. Microsoft claimed it could compete with Sonnet 4.6 and Opus 4.6 in coding, according to human evaluations. However, the benchmarks published at the time revealed a different reality, with Thinking-1 significantly lagging behind the competition from OpenAI and Anthropic, ranking roughly on par with Deepseek V3.2.
Customers May Get Lower-Performing AI for the Same Price
For Copilot and Office customers, this could mean paying the same amount for lower-performing AI so that Microsoft can reduce its own costs. Mustafa Suleyman, head of AI at Microsoft, openly acknowledged this plan in June: "We spend a lot of money on Anthropic—so our goal is to reduce and ultimately eliminate that cost." This is the same Microsoft that recently argued that reliance on providers like OpenAI and Anthropic is a bad thing and that it wants to be a neutral platform alternative.
Microsoft CEO Satya Nadella also hinted that AI billing could evolve toward a usage-based pricing model rather than fixed-rate subscriptions. A possible setup could make the cheaper MAI models the default choice, while third-party models from OpenAI or Anthropic would be available as premium options at an additional cost. Thus, Microsoft would pass its expenses related to OpenAI and Anthropic onto customers with a markup.
Microsoft also stated that the MAI models are trained on proprietary and commercially licensed data, making them safe for businesses. However, according to the technical document, Microsoft has used the Common Crawl dataset, a collection of freely accessible web data, the legality of which for AI training is not well established. Every other AI company does the same, but Microsoft presents its own training data as particularly clean.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.