Microsoft's AI model division, MAI, announced a family of seven new models at Microsoft Ignite 2026, built around a single pitch: frontier-class capability at a price developers can afford to run agents on continuously. "As token spend skyrockets, we know that you also want models at a price point that lets you run agents 24/7 doing real engineering work," the company said in its announcement.

The launch, covered widely by the developer community, positions Microsoft's in-house models as direct price-performance challengers to OpenAI and Anthropic — and adds a new licensed-weights program that lets enterprises run customized MAI models inside their own Azure environments. For ongoing coverage of the latest AI developments from every major lab, AI Buzz Wire tracks each release as it lands.

The Coding Models: Frontier Performance, Discounted

The headline claim concerns MAI Code 2, built for everyday enterprise coding tasks. Averaged over four widely used benchmarks that target the work seen in GitHub Copilot — DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1 — Microsoft says MAI Code 2 delivers 98% of the performance of GPT-6.1 Sol at 40% of the price, and 96% of Claude Opus 5.5's performance at just 5% of the price.

In a blind, side-by-side evaluation of frontend work, developers preferred MAI Code 2 over GPT-6 Sol 64% of the time, according to Microsoft, which also said the model has been tuned to generate dashboards and visualizations. The model is available in GitHub Copilot today. Microsoft was explicit about limits, noting the models are "explicitly not yet tuned for harder problems, like fundamental science research and very long horizon agentic work."

A smaller sibling, MAI Code 2 Flash, targets high-volume, everyday tasks like code explanations, targeted edits and routine fixes. Across three coding benchmarks measuring routine work, Microsoft says it performs comparably to GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of their prices respectively. It is available in GitHub Copilot and in a limited-access preview in Microsoft Foundry.

Cyber Defense at 2% of the Cost

MAI Cyber 2 Flash extends the company's security-focused model line, which Microsoft says already accounts for almost 40% of scans in MDASH, its security operations platform, with 90% of scans that start with MAI continuing to use it. The new model is priced 86% lower on output tokens than MAI Cyber 1.1 Flash, giving existing customers a stronger model at a lower price.

"Security, alignment and containment are the most important priorities for agent development in 2027 and beyond," the company said, describing a future where monitoring and control agents defend and patch live production systems around the clock.

Image, Voice and Transcription Models

MAI Image 2.7 takes the number one spot on Artificial Analysis for image editing, outperforming GPT Image 2.6 Sunburst (max) at one-fifth of the price, according to Microsoft, which ranked fourth in text-to-image generation — making it the second-best lab by that measure. MAI Image 2.7 Flash brings high-volume generation at roughly $10 per 1,000 images. Both are available in Microsoft Foundry.

MAI Realtime 1 is Microsoft's first live speech-to-speech model, designed for natural, low-latency conversations. The company says it outperformed GPT-Realtime 2.1 across external and Dynamics 365 benchmarks, with pricing starting at $0.01 per audio minute. It can be tried in MAI Labs.

On the medical side, Dragon Copilot Transcribe, built on MAI Transcribe 2, is now live and powers 100% of doctor-patient conversation transcriptions in Dragon Copilot, covering 10 million encounters a month, halving error rates in multilingual workflows, and used by a third of US physicians, according to the announcement.

Cheaper Routing, Under Your Control

Rounding out the lineup, Decision 1.1 provides model routing built into applications so that larger, pricier models are only invoked when needed. Microsoft says it leads the Artificial Analysis Decision Index ahead of Jev, Google and OpenAI, is used by 4,000 customers, and is 50 times faster and 79 times cheaper than GPT-6.1 Sol.

The most strategically significant announcement may be MAI Licensed Weights. Enterprises can take a frontier MAI model, customize it with their proprietary data, and run it securely in their own Azure environment, controlling access while keeping their data and adaptations. Microsoft framed the program as an answer to enterprises wary of building on proprietary APIs or open weights of unknown provenance: "Your IQ. Your agents. Your weights. All under your control."

The company pointed to AT&T as an early example: the telecom giant developed an MAI custom model for 5G network operation analytics that Microsoft says outperformed GPT-6.1 Sol at one-fiftieth of the cost.

A Price War Window Opens

The Ignite announcements arrive amid rising scrutiny of AI spending across the industry, and Microsoft is clearly betting that unit economics — not raw benchmark supremacy — will decide enterprise procurement in the next cycle. With claims like Claude Opus 5.5-level performance at 5% of the price and state-of-the-art image editing at a fifth of the cost, the message to rivals is blunt: the frontier is getting cheaper.

Whether independent benchmarks confirm Microsoft's numbers remains to be seen, but the scale of the rollout — seven models spanning code, cyber, image, speech and transcription, all day-one available — signals that MAI intends to compete across the full stack rather than in a single niche.

Stay Ahead of AI

Model releases are accelerating across every major lab. Keep up with the latest AI developments, benchmarks and price moves at AI Buzz Wire.

Read more AI news →