Microsoft launches seven cheaper in-house AI models at Ignite

Microsoft unveiled seven new MAI models at Ignite, including coding and cyber-defence systems it says match rival quality at a fraction of the price.

অনুবাদ চলছে… 0 / 4 মূল লেখা পড়ুন

Satya Nadella on stage in front of the Microsoft logo at Ignite 2026
Microsoft chief executive Satya Nadella speaks at Ignite 2026 in San Francisco. — MICROSOFT

— Microsoft unveiled a family of seven new artificial intelligence models at its Ignite conference on Tuesday, 17 November, saying they deliver close to the quality of leading rival systems at a small fraction of the cost.

The models come from Microsoft AI, the company’s in-house model unit, which says customers want to run software agents around the clock doing real engineering work as spending on AI tokens climbs. “Our mission is to build you the very best models in the world,” the company said in its announcement, adding that cost is now as important as capability.

Coding and cyber defence

MAI Code 2, the flagship coding model, reaches 98 per cent of the quality of OpenAI’s GPT 6.1 Sol at about 40 per cent of the price, Microsoft said. Against Anthropic’s Claude Opus 5.5 it reaches 96 per cent of the performance at about 5 per cent of the cost. Those figures are averaged over four coding benchmarks that Microsoft says mirror everyday enterprise work in GitHub Copilot.

A lighter version, MAI Code 2 Flash, is built for high-volume jobs such as code explanations and routine fixes. Microsoft says it matches OpenAI’s GPT-6 Luna at 65 per cent of the cost and Anthropic’s Claude Haiku 5.5 at 20 per cent of the cost. A security model, MAI Cyber 2 Flash, is said to beat OpenAI’s GPT-6 Astra on the CyberGymE2E benchmark at less than 2 per cent of the cost per task.

Images, voice, and chips

The rest of the lineup covers other media. MAI Image 2.7 is ranked first on the Artificial Analysis image-editing leaderboard and costs about one-fifth as much as OpenAI’s GPT Image 2.6 Sunburst, Microsoft said. MAI Realtime 1, a speech-to-speech model, starts at $0.01 per audio minute. MAI Transcribe 2 underpins Dragon Copilot Transcribe, which Microsoft says already handles 10 million patient encounters a month and has halved multilingual error rates. The seventh model, Decision 1.1, is described as 50 times faster and 79 times cheaper than GPT-Sol 6.1.

Microsoft also said its second-generation Maia 200 AI accelerator is now in production running its largest models, with 1.7 times more tokens per watt.

What the numbers do not show

The performance and price comparisons are Microsoft’s own, and have not been independently verified. The company also acknowledged limits: the coding models are “explicitly not yet tuned for harder problems, like fundamental science research and very long horizon agentic work”.

TT Foreign Desk

World news from The Thursday Times' foreign desk.

More from this desk

The conversation

Comments are open on this story. An editor reads every one before it appears.

No one has said anything yet. Be the first.

Published by Thursday Media International thursday.media