AI NewsModels & agentsAnnouncement

Microsoft launches seven in-house AI models at Ignite 2026

Microsoft announced seven in-house AI models at Ignite 2026 covering code, cyber defence, images, speech and model routing.

AI News

Editorial2 min read

LinkedInX
Signal-field composition representing a Microsoft model launch.

Why it mattersMicrosoft now sells its own first-party models that price against Anthropic and OpenAI. A buyer picking a coding model has more price options this week.

Microsoft is no longer relying only on its OpenAI partnership for AI. At Ignite on 17 November 2026, the company announced seven in-house MAI models. Every one of them is priced below a competitor Microsoft names.

The seven models cover coding, cyber defence, image editing, speech, medical transcription and model routing. Microsoft also announced the Maia 200 accelerator. The company says the chip runs at 1.7 times more tokens per watt than its predecessor. Every price and benchmark below is Microsoft's own, stated on the Microsoft AI news page.

The two coding models and where they land

MAI Code 2 is for everyday enterprise coding. Microsoft says it reaches "98 percent of the coding quality of GPT-6.1 Sol" at about 40 percent of the price. Against Claude Opus 5.5, Microsoft says the model hits 96 percent of the quality for 5 percent of the price. In internal frontend work, developers preferred MAI Code 2 over GPT-6 Sol 64 percent of the time.

MAI Code 2 Flash is the cheap tier. Microsoft compares it to GPT-6 Luna at 65 percent of the price. The same model is compared to Haiku 5.5 at 20 percent of the price. Both coding models are in GitHub Copilot now. Flash is in limited preview in Microsoft Foundry.

The cyber, image, speech and routing models

MAI Cyber 2 Flash is for cyber defence. Microsoft says it "beats GPT-6 Astra's quality" at less than 2 percent of the cost per task. It is already running inside MDASH, which is Microsoft Defender's own AI tool. The model is 86 percent cheaper on output tokens than MAI Cyber 1.1 Flash.

MAI Image 2.7 is first on the Artificial Analysis image editing board, Microsoft says. Its Flash variant costs about $10 per 1,000 images. MAI Realtime 1 is a live speech-to-speech model. It starts at $0.01 per audio minute. Microsoft says it beats OpenAI's GPT-Realtime 2.1. Dragon Copilot Transcribe is for medical conversations. Microsoft says it "halves error rates in multilingual workflows." Dragon Copilot already powers 10 million patient encounters a month.

Decision 1.1 is a routing model that picks the right model for each task. Microsoft says it reaches general availability with 4,000 customers. The company states the router is "50x faster and 79x cheaper than GPT-6.1 Sol" for the routing decision itself.

What a buyer should check

Every figure above is a vendor benchmarking itself against a named rival. The reader's job is to test the claim on their own work. A team using GitHub Copilot for routine edits can try MAI Code 2 on real pull requests. A team buying image editing can test MAI Image 2.7 Flash on 1,000 of its own files. A team that routes between models can run Decision 1.1 against a traffic sample and read its own bill.

Microsoft's position inside the OpenAI partnership has not changed. What is new is that Microsoft sells models that compete on price with the model its partner sells. The place a developer or enterprise buyer meets them is familiar: GitHub Copilot and Microsoft Foundry.

Source

SourceMicrosoft

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project