Skip to content
DealsBlack Friday & Cyber Monday · 19-30 November · tech deals, re-checked every hourSee the deals
Newsletter
News

Microsoft Unveils Seven In-House AI Models at Ignite, Pitched on Price

Microsoft's MAI family spans coding, cybersecurity, images, speech, transcription and decision-making, and the company's own benchmarks claim near-rival performance for a fraction of the cost.

Microsoft Unveils Seven In-House AI Models at Ignite, Pitched on Price

What you need to know

  • Microsoft announced seven MAI models and model-powered services at Ignite on 17 November 2026, covering coding, cybersecurity, images, live speech, transcription and automated decisions.
  • MAI Code 2 and Code 2 Flash are available in GitHub Copilot now. Microsoft claims Code 2 delivers 98% of GPT-6.1 Sol's coding performance at about 40% of the price, though that figure uses a 50% promotional price.
  • All prices are in US dollars. UK pricing in pounds has not been confirmed for any of the models.

Microsoft has unveiled seven AI models and model-powered services built by its own Microsoft AI division. They were announced on 17 November at its Ignite conference, which runs in San Francisco and online until 20 November. The company describes the release as "a family of seven new world class models". The models are aimed at coding, cybersecurity, image generation, speech, transcription and automated decision-making, and they are pitched as cheaper, specialised alternatives to rival systems.

The two coding models, MAI Code 2 and MAI Code 2 Flash, went live in GitHub Copilot on the day of the announcement.

The seven models

  • MAI Code 2 handles complex, agentic coding.
  • MAI Code 2 Flash handles faster, high-volume coding.
  • MAI Cyber 2 Flash is built for cybersecurity.
  • MAI Image 2.7 covers image editing and generation.
  • MAI Realtime 1 runs live speech-to-speech conversations.
  • Dragon Copilot Transcribe provides multilingual transcription and is built on MAI Transcribe 2.
  • Microsoft Decision 1.1 handles routing, classification, prioritisation and verification.

Coding: the headline claims

Microsoft says MAI Code 2 is designed for everyday enterprise coding. That includes deep debugging, multi-file changes, repository-wide reasoning and interface work.

The company tested it across four benchmarks: DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1. On those, Microsoft says MAI Code 2 achieved:

  • 98% of GPT-6.1 Sol's coding performance at about 40% of its price
  • 96% of Claude Opus 5.5's performance at 5% of its price

Microsoft also says developers preferred MAI Code 2 over GPT-6 Sol 64% of the time in blind frontend tests involving 258 annotators.

There are two important caveats:

  • The price comparisons use a 50% promotional price for MAI Code 2.
  • Microsoft concedes the model is not yet tuned for harder work such as fundamental scientific research or very long-horizon agentic tasks.

Code 2 Flash targets quicker jobs such as code explanations, targeted edits and routine fixes. Across three of the same benchmarks, Microsoft says it is comparable to GPT-6 Luna at 65% of the price and to Claude Haiku 5.5 at 20% of the price. Those comparisons also use the promotional rate.

In GitHub Copilot, the two models are available on different plans, both subject to eligibility and usage-based billing:

  • Code 2: Pro, Pro+, Max, Business and Enterprise
  • Code 2 Flash: Free, Student, Pro, Pro+, Max, Business and Enterprise. It is also in limited-access preview in Microsoft Foundry.

Security, images and speech

MAI Cyber 2 Flash. Microsoft says this model matches or beats GPT-6 Astra's quality on the CyberGym-E2E benchmark at less than 2% of the cost per task. It is also priced 86% lower on output tokens than its predecessor, MAI Cyber 1.1 Flash. For now it is in preview only within Codename MDASH, Microsoft's agentic code-security system. According to the company, MAI Cyber models already handle almost 40% of MDASH scans.

MAI Image 2.7. Microsoft says it ranked first for image editing on Artificial Analysis, beating GPT Image 2.6 Sunburst Max at one-fifth of the price. It placed fourth for text-to-image generation. A high-volume sibling, MAI Image 2.7 Flash, costs approximately $10 per 1,000 images. Both are offered through Microsoft Foundry and are labelled as previews.

MAI Realtime 1. This is Microsoft's first live speech-to-speech model. The company says it outperformed GPT-Realtime 2.1 in its evaluations. Pricing starts at $0.01 per audio minute, and it can be tried in MAI Labs. Broad commercial availability has not been confirmed.

Transcription and decisions

Dragon Copilot Transcribe is already running inside Microsoft's Dragon Copilot healthcare product rather than being a new consumer app. Microsoft says it:

  • powers 10 million patient encounters a month
  • has halved multilingual error rates
  • is used by approximately one-third of US physicians

It builds on MAI Transcribe 2 Streaming, announced on 1 October. That model supports real-time transcription in 60 languages at an introductory $0.54 per audio hour until the end of 2026.

Microsoft Decision 1.1 is generally available in Microsoft Foundry, OpenRouter and Vercel. It is built on Gemma 4 26B A4B and makes quick calls inside software and agents. Microsoft already uses it to route requests in Copilot.

Microsoft's own figures for Decision 1.1 do not line up. The company's MAI announcement says it is 50 times faster and 79% cheaper than GPT-6.1 Sol. In his keynote, chief executive Satya Nadella said it was:

more than 10x faster and 200 times less expensive than using an LLM

The two claims appear to use different baselines.

What's next

Microsoft is also offering MAI Licensed Weights to selected partners. This lets them customise its models with their own data and deploy them in their own Azure environments. Pricing, eligibility numbers and any wider release date have not been confirmed.

Every price announced so far is in US dollars. UK pricing in pounds has not been confirmed for any of the new models.

Why it matters

For UK developers already paying for GitHub Copilot, this is the most immediate change: a new Microsoft-built coding model sits in the model picker, and Code 2 Flash is listed even for Free and Student users, subject to eligibility and usage-based billing. The headline cost savings are Microsoft's own figures, and several rely on a 50% promotional rate, so the real price gap once the promotion ends is unclear. More broadly, Microsoft is now benchmarking its own models directly against OpenAI's and Anthropic's. More competition among model makers is likely to put downward pressure on the cost of the AI features built into the apps and services people use.

Sources and evidence (9)
↑↓ to moveEnter to openEsc to close
GadgetScout

What do you need?

Browse by gadget

Topic hubs

Shop by category

More