Microsoft Launches Decision 1.1 and Expands Its AI Range

At Microsoft Ignite 2026, Microsoft launched Decision 1.1 and expanded its range of MAI models. Among them, MAI-Code-2 is priced at the standard rate of $1 per million tokens for input and $5 per million for output.
Companies Still Need to Validate Performance
For Mitch Ashley, an analyst at Futurum Group, the question is whether the models provide an advantage to customers in the products they are already using. In his view, Microsoft can afford not to dominate the competition among standalone models if GitHub Copilot, Foundry, and its security products offer useful features at better economic conditions.
The analyst specifically calls for independent validation by companies of the performance of Cyber 2 Flash, which he describes as a low-cost specialized model for agent security. A reduced price is not always preferable if an attack goes unnoticed or if an inappropriate fix is triggered, he warns. He applies this caution to Code 2 and Code 2 Flash as well.
According to Mitch Ashley, the relevance of the price must be examined on an application-by-application basis. Companies will assess the quality of the outcome and the overall expenditure required to achieve it, not just the token billing.
Foundry Chooses Models and Microsoft Displays Its Prices
Foundry, Microsoft's platform dedicated to the large-scale creation and deployment of AI applications and agents, has a router that automatically selects the model suited for the task. Decision 1.1 relies on Gemma from Google DeepMind, whose weights are open. It ensures fast routing, which Microsoft describes as cost-effective, as well as tasks of classification, prioritization, and verification. Developers can provide it with instructions to guide its decisions based on the requested work.
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.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Michael Heath, principal cloud and AI architect at Myriad360, a global Microsoft partner integrator, believes that his clients are primarily looking for the ability to replace a model without having to rebuild their agents. According to him, automatic selection helps them prioritize outcomes over technical choices or expense monitoring. He considers the MAI range to be more mature, competitive, and less expensive, and sees Microsoft's choices as prioritizing performance-to-price ratio over just the capabilities of the models.
In comparison to the pricing of MAI-Code-2, OpenAI's GPT-6.1 is offered at $2 for input and $10 for output per million tokens. For the same billing unit, Anthropic's Claude Opus 5.5 lists $4 and $20, respectively.
MAI Models for Code, Image, and Security
For programming, MAI-Code-2 targets complex tasks assigned to agents, while MAI-Code-2-Flash is aimed at routine tasks processed in large volumes. These models are accessible on GitHub. The Flash version is also offered in Foundry as a private preview.
On the image side, MAI-Image-2.7 and MAI-Image-2.7-Flash are available in preview, aiming for more precise retouching and consistent image production. Also accessible in preview, MAI-Cyber-2-Flash assists security teams in spotting, confirming, and correcting vulnerabilities.
Microsoft plans to reserve priority access to MAI weights for partners that the company designates as "trusted." On Tuesday, during his conference speech, Satya Nadella, Chairman and CEO of Microsoft, expressed his desire to enable everyone to own their own intelligence.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.