FRIDAY, NOVEMBER 20, 2026 Tech news, minus the noise.
N News Cast
LATEST
Home  /  AI
AI

Microsoft's Seven New MAI Models Are a Bet That Cheaper AI Wins

Microsoft unveiled seven specialized AI models at Ignite 2026, led by a coding model it says matches 98 percent of GPT-6.1 Sol at 40 percent of the price. The pitch is simple: stop paying frontier prices for routine work.

Microsoft's Seven New MAI Models Are a Bet That Cheaper AI Wins
Source: Gangulybiswarup via Wikimedia Commons (CC BY 4.0)

Microsoft unveiled seven specialized AI models at its Ignite 2026 conference in San Francisco, and the message was blunt: stop paying frontier prices for routine work. The Microsoft MAI models cover coding, cyber defense, image generation, live voice, transcription and model routing, and Microsoft says the coding model does 98 percent of the work of GPT-6.1 Sol at roughly 40 percent of the price. The pitch is not that these models are smarter. It is that they are cheap enough to use everywhere.

That is a different kind of AI announcement. Most model launches are about capability: bigger benchmarks, longer context, harder reasoning. This one is about the bill. Microsoft is betting that the next wave of AI spending goes to models that are good enough for the job at hand, not the most powerful model in the catalog.

What the Microsoft MAI models actually do

The lineup splits the work into seven jobs. MAI Code 2 is the coding model, aimed at multi-file changes and bugs that take real digging. MAI Code 2 Flash is the lighter, faster version, and GitHub says it is available to Copilot Free and Student users as well as paid customers. MAI Cyber 2 Flash handles security triage, the kind of alert sorting that eats security teams alive.

MAI Image 2.7 and a Flash variant cover image generation and editing, and both are in public preview on Foundry. MAI Realtime 1 is Microsoft's first live speech-to-speech model, priced from one cent per audio minute. MAI Transcribe 2 does transcription, and Decision 1.1 routes work to the right model instead of sending everything to the most expensive one.

None of these is meant to headline a keynote on its own. Together they cover the unglamorous bulk of AI usage: the code fixes, the meeting transcripts, the image edits, the voice calls. That is the point. Microsoft is selling a fleet, not a flagship.

The price is the product

The number Microsoft wants you to remember is 98 percent at 40 percent of the price. That is its claim for MAI Code 2 against GPT-6.1 Sol across four coding benchmarks. Treat it as a vendor number until independent tests weigh in, but the direction is clear. Microsoft is competing on cost per task, not just on capability.

This fits a pattern that has been building all month. Gemini 4 Argon arrived with a focus on practical deployment, and GPT-6 rolled out to everyone with a friendlier interface. Every lab now has a cheaper tier, and every lab is learning the same lesson: the model that wins is often the one the finance team approves, not the one that tops the leaderboard.

Microsoft is not abandoning frontier models either. GPT-6.2 is coming to Foundry alongside the MAI lineup. The strategy is a shelf with two levels: frontier models for the hardest work, MAI models for everything else. The question for buyers is how much of their workload actually belongs on the top shelf.

Why this matters beyond Microsoft

When the biggest enterprise software company in the world starts selling AI by the price tag, the market follows. The AI agent wars have mostly been fought over who can build the smartest agent. The next phase will be fought over who can run agents cheapest at scale, because agents that cost too much per task never get deployed.

For developers and IT buyers, the practical effect is choice. A coding assistant that costs 40 percent less changes what you automate. A transcription model at a fraction of the old price changes which meetings get recorded. A routing model that picks the cheap option by default changes the monthly bill without anyone noticing.

What to watch

Three things will decide whether this launch matters. First, independent benchmarks. Microsoft's 98 percent figure comes from its own testing, and coding benchmarks are easy to flatter. Watch what happens when the models meet messy, real-world codebases.

Second, availability. The image models are in public preview, and enterprise rollouts have a habit of arriving slower than the keynote promises. Third, pricing discipline. Cheap tiers have a way of getting more expensive once everyone depends on them. If Microsoft holds the line, the MAI models could reset what companies expect to pay for everyday AI. If not, they are just another launch-day discount.

AS
News Cast Editorial Team

We cover AI, gadgets, startups and security — original reporting and explainers, written by humans, for humans. Got a tip? Reach us via the contact page.