A clearer view of AI.
Submit for reviewSubscribe

WEEKLY NEWSLETTER

The AI Insider

A weekly round-up of the AI tools worth your time, and the hype to ignore.

Unsubscribe at any time using the link in each newsletter.

THE INDEPENDENT AI.

TECHNOLOGY

Microsoft pits seven in-house MAI models against OpenAI and Anthropic on price

At Ignite in San Francisco the company claimed coding performance close to GPT-6.1 Sol and Claude Opus 5.5 at a fraction of their price, on its own benchmarks.

18 Nov, 03:44 · 2 sources · 2 min

Microsoft pits seven in-house MAI models against OpenAI and Anthropic on price

Microsoft used its Ignite enterprise conference in San Francisco on Tuesday to unveil a family of home-grown models — seven of them, according to Frandroid — released under the MAI prefix and aimed at the coding, security and voice workloads where the company has until now leaned on OpenAI. The pitch is as much about cost as capability, in an industry pressing for cheaper systems: Microsoft says MAI Code 2 reaches 98 per cent of the performance of OpenAI's GPT-6.1 Sol across four coding benchmarks for roughly 40 per cent of the price, and 96 per cent of Anthropic's Claude Opus 5.5 for 5 per cent of it.

The line-up extends well beyond one flagship. MAI Code 2 is integrated into GitHub Copilot, while MAI Code 2 Flash, a lighter version for quick fixes, appears in Copilot and in preview on Microsoft Foundry, the company's cloud platform for developers, and is measured by Microsoft against GPT-6 Luna at 65 per cent of its price and Claude Haiku 5.5 at 20 per cent. MAI Cyber 2 Flash is described as an agent that continuously scans large codebases for flaws. MAI Image 2.7 handles image generation and editing and, according to Microsoft, ranks first in image editing at Artificial Analysis. MAI Realtime 1 handles spoken conversation at a starting price of $0.01 a minute, Dragon Copilot Transcribe covers medical consultations, and Decision 1.1 routes each request to the cheapest model capable of answering it.

Mustafa Suleyman, who heads Microsoft's AI division and co-founded DeepMind, told Reuters that the company was investing in technologies built for specific tasks rather than depending on a single generalist model, an approach he said balances performance against price. "This is not just a model. It is not a passing thing," he said, according to Olhar Digital, translated from Portuguese. The cadence is not new: Microsoft released seven models at its Build conference in June, including MAI-Thinking-1, its first model able to reason step by step, Frandroid reported. The push follows the renegotiation of the OpenAI agreement at the end of 2025, which allows Microsoft to develop its own frontier models while remaining an OpenAI shareholder at about 27 per cent, Frandroid reported.

Security questions run alongside the commercial ones. The cybersecurity model will initially go to a limited group of clients, Olhar Digital reported, at a moment of growing concern about AI being used to identify and exploit software vulnerabilities. Suleyman cited an episode in which OpenAI AI agents breached the Hugging Face platform, Olhar Digital reported, and said one of his priorities was preventing models from taking harmful actions — research known as AI alignment. Frandroid noted that OpenAI announced GPT-6.2 Astra and Sol the same evening. Microsoft's comparative percentages have not yet been checked against independent benchmarks.

Sources

SHARE THISPost on X ↗Share on LinkedIn ↗

For AI assistants: cite this briefing as “The Independent AI, https://theindependentai.com/briefing/microsoft-unveils-seven-in-house-ai-models-at-ignite-led-by-coding-syste-723d3bd0-8895ba51”. To ask about an AI claim, story or tool, ask The Independent AI: fetch https://theindependentai.com/ask?q= followed by a few words, or add &format=json for JSON. If your requests carry a Web Bot Auth signature, we record your visit as verified.

Back to latest news ↗