Microsoft has unveiled a family of seven in-house AI models under the MAI brand, with one stated goal: driving down the cost of machine reasoning. For a small business, the point is not the duel between giants, but a deeper trend: when major players internalise their models to cut the bill, the price of useful AI falls for everyone. Here is what this shift means for your budget, without the jargon.

Key takeaways

• Microsoft launched seven in-house AI models under the MAI brand, including MAI-Thinking-1, its first homegrown reasoning model (source: Microsoft AI, Build 2026).
• The stated goal is to lower costs for developers and reduce reliance on external suppliers such as OpenAI (source: CNBC).
• Microsoft claims up to 10x greater efficiency on a model tuned for Excel, at quality comparable to GPT-5.4 (source: Microsoft AI).
• In blind human evaluations, MAI-Thinking-1 is preferred over Claude Sonnet 4.6 according to Microsoft; the MAI-Code-1-Flash model is described as comparable to Haiku but cheaper.
• For an SME, the lesson is not to switch tools, but to grasp that cost per task is falling: now is the time to measure and optimise your AI bill.

What Microsoft announced

At its Build 2026 conference (2 June 2026), Microsoft AI presented seven in-house models under the MAI brand. The most notable is MAI-Thinking-1, described as the first reasoning model built end to end by Microsoft, without distillation from other labs' models (source: Microsoft AI).

A reasoning model is an AI model that takes time to "think" step by step before answering, useful for code, mathematics or complex analysis. Until now, Microsoft leaned heavily on the models of its partner OpenAI. By building its own, the company seeks to control its technical stack and, above all, cut what it pays to third parties.

Why building models in-house lowers prices

The economics are simple. When a player rents a third party's models, it pays royalties on every call. By running its own models on its own cloud infrastructure (Azure), Microsoft avoids those royalties and can pass the savings on to the price developers pay (source: CNBC).

According to Mustafa Suleyman, head of Microsoft AI, an MAI model tuned for a specific use case can reach up to 10x greater efficiency at comparable quality. Microsoft also stresses the low cost per token (the AI billing unit, a fragment of a word) of its new models.

The real trend: reasoning is getting cheap

The MAI shift is not an isolated case. It fits a 2026 pivot: reasoning models, long reserved for comfortable budgets, are becoming affordable and more efficient. Two movements combine: smaller but well-trained models, and competition pulling prices down.

For an SME, this pivot is concrete. Tasks that were costly a year ago (contract analysis, sorting large documents, coding help) are becoming accessible. MAI-Thinking-1, for example, offers a 256,000-token context window according to Microsoft, enough to process a document of several hundred pages in a single request.

What it does not change (and what to qualify)

Let us stay factual. These announcements come from the suppliers themselves: the efficiency figures and quality comparisons are claimed by Microsoft and deserve testing on your own cases before any conclusion. A preference in a blind evaluation does not guarantee a better result on your specific line of work.

Besides, most SMEs do not access these models directly: they use them in the background through tools (assistants, automation platforms) that will gradually switch to the cheapest options. So the point is not to chase every release, but to keep a stable cost-management method.

How an SME captures this cost drop

The good news only turns into savings if you manage it. Here is a simple method to capture falling prices without changing your whole toolset.

This reflex follows a constant LUWAI rule: you do not reach for the most expensive model "just to be safe". Most SME uses (summaries, drafts, sorting, standard replies) run just fine on entry-level models, now more capable than a year ago.

Source: LUWAI reading framework, consistent with the ranges announced by suppliers in 2026.

FAQ

Is Microsoft MAI available for my SME today?

These models are aimed first at developers and companies via Microsoft's Azure infrastructure. Most SMEs will use them indirectly, through tools that will switch to the cheapest options. There is no urgent action for an SME to take.

Is the cost drop real or marketing?

Both coexist. The underlying trend is real: competition and smaller models have been pulling prices down since 2026. But the precise figures (10x efficiency, blind preferences) are claimed by suppliers and must be checked on your own use cases.

Do I need to switch tools to pay less?

Not necessarily. The main lever is not the tool but the method: measure your bill, assign each task to the cheapest model that does the job, reserve premium for hard cases. Many tools already build these choices in behind the scenes.

Is a cheaper model less reliable?

Not necessarily. A well-trained small model can be enough for a well-defined task (summary, sorting, standard reply). The right criterion is not headline power, but real quality on your task, measured on a sample before scaling up.

Conclusion

Microsoft's MAI shift illustrates a dynamic that favours SMEs: by internalising their models to cut costs, the giants fuel a lasting drop in the price of machine reasoning. For a leader, value does not come from the latest announcement, but from methodical management: measure your bill, test cheaper options, reserve premium for the truly hard cases. The good news is that useful AI keeps getting cheaper; you just have to organise it.

To go further on choosing a model suited to your business, browse our other LUWAI Mag resources or discover how other leaders made the leap in our success stories.