Microsoft claims its own MAI models match rivals at a fraction of the price

The company is releasing MAI Code 2 and a cybersecurity model at its Ignite conference, claiming they nearly match rivals' models at a fraction of the price. At the same time, Copilot still routes work to competitors' models.

File photo: Building 92 at Microsoft Corporation headquarters in Redmond, Washington. Photographed by user Coolcaesar on 30 May 2016.
File photo: Coolcaesar, CC BY-SA 4.0 · source This adaptation is licensed under CC BY-SA 4.0.

Microsoft claims its own MAI models match rivals at a fraction of the price

The company is releasing MAI Code 2 and a cybersecurity model at its Ignite conference, claiming they nearly match rivals' models at a fraction of the price. At the same time, Copilot still routes work to competitors' models.

Microsoft on Tuesday, November 17, 2026, launched a series of AI models with strong coding and cybersecurity capabilities, in a move that puts the company in direct competition with Anthropic and OpenAI. The launch took place at the company's Ignite conference in San Francisco and includes the coding model MAI Code 2 and a model built specifically for cybersecurity, according to Reuters.

What was announced

MAI Code 2 is, according to Microsoft, a code generator that "almost matches" competitors' models on performance, but at a fraction of the cost. The numbers behind the claim have not been made public, and no independent source has verified them — this is the company's own assessment, relayed through Reuters.

The cybersecurity-specific model will first be offered to a limited group of customers. The reason, according to Reuters, is an increase in software vulnerabilities that AI tools are increasingly surfacing. The model is not named in the Reuters article, but The Verge writes, citing Axios, that it is called MAI Cyber 2 Flash — a discrepancy we cannot resolve further here. Axios further states, according to The Verge, that the launch encompasses six models in total; this too is a reported figure we cannot verify.

In addition, Microsoft launched models for image generation and audio transcription, according to CNBC TV18 with contributions from Reuters. The launches build on the company's in-house MAI model family, part of the effort to reduce dependence on external AI providers.

Benchmark claims and a "hill-climbing machine"

Microsoft's AI chief Mustafa Suleyman told Reuters that the strategy of domain-specific models keeps costs down while, in his view, matching competitors and in some cases beating them on benchmarks. One claim stands out in particular: According to The Verge (via Axios), Suleyman claims that MAI Cyber 2 Flash beats OpenAI's GPT-6 Astra on the CyberGymE2E benchmark. This is a company-sourced claim without independent verification in the sources.

In the interview, Suleyman described the work not as individual models but as a system:

"This isn't just one model. It's not a flash in the pan," he told Reuters. "This is a system, a hill-climbing machine that we have built, which co-optimizes performance and price."

The reason for the statement is that, according to Reuters, he believes domain specialization makes it possible to compete with much larger general models in the areas that matter most to Microsoft's customers.

Competing with the rivals — and distributing them

The awkward part of the picture is that Microsoft's own Copilot platform still rests on competitors' models. In the Copilot blog, the company writes that "Copilot brings together leading models from across the industry. And with Auto routing, employees don't have to worry about choosing the right model for each task" (Microsoft Copilot Blog). The Auto routing is said to balance quality, speed, and cost — features that also mean OpenAI and Anthropic models continue to get work done through Microsoft's platform.

So this is not a break with its partners, but a strategy in which Microsoft tries to cut costs by replacing external models with in-house ones where they are "good enough" at a lower price. The existence of the MAI family suggests this work has been going on for some time; the Ignite launch is its most visible marker so far.

The security focus

Suleyman also told Reuters that a recent cybersecurity reversal — illustrated by OpenAI agents reportedly having hacked the Hugging Face platform — has led him to focus not only on performance but also on critical safety mechanisms, and that alignment is his highest priority. The details of the Hugging Face episode are not documented in the sources underlying this article, so we report it as Suleyman describes it to Reuters.

The reason the cybersecurity model is first getting a limited rollout is likely both commercial and cautious: the sector has seen an increase in vulnerabilities surfaced with AI tools, and a new security tool will naturally be tested before broad deployment.

Open questions

A fair amount remains before we know what the launch actually means:

  • Price and performance are unverified. What "a fraction of the cost" amounts to in actual figures, and how "almost matches" holds up in independent testing, has not been made public.
  • The CyberGymE2E claim that MAI Cyber 2 Flash beats GPT-6 Astra comes from Suleyman himself. The benchmark has not been established as a common industry standard in the sources.
  • The name of the cybersecurity model is uncertain: MAI Cyber 2 Flash is noted by The Verge via Axios, while the Reuters article does not name it.
  • The figure of six models comes only from Axios via The Verge.
  • When the cybersecurity model becomes available to more than the first selected customer group is not specified.

The launch otherwise took place at an Ignite keynote where Satya Nadella spent much of his time on something else: apologizing for the global Outlook outage the previous week, which affected users in more than 70 countries. That matter is separate from the model launch, but it framed the conference.

AIMag.no
AIMag.no
The AIMag.no editorial team covers artificial intelligence, tools, research, and regulation.

Get the best of AI MAG in your inbox

News, analysis, and ideas at the intersection of AI and society.