
Key Points
- 01Microsoft unveils six MAI models for coding, security, images and voice at Ignite
- 02MAI Code 2 claims near‑rival performance at sharply lower prices
- 03MAI Cyber 2 Flash targets cybersecurity with major cost savings per task
- 04Microsoft also pursues a frontier general‑purpose model and offers model weights to enterprises
Microsoft expands MAI lineup with six new models
Microsoft introduced six new MAI artificial intelligence models on Tuesday at its Ignite conference in San Francisco, broadening its in‑house lineup across coding, cybersecurity, image generation, transcription and real‑time voice. The launches are aimed at competing more directly with leading systems from OpenAI and Anthropic while emphasizing lower running costs. The new releases extend a MAI family that already covers reasoning, coding, images, transcription and voice.
The models sit under the Microsoft AI, or MAI, brand and target four main categories of work: writing code, defending systems, generating images and producing or processing voice. Cybersecurity is highlighted as a newer focus within the MAI family, which previously concentrated on areas such as reasoning and coding. The announcements mark a further step in Microsoft’s effort to rely more on its own models rather than depend entirely on external AI providers.
Pricing and performance claims against OpenAI and Anthropic
Microsoft says MAI Code 2, its new programming model, delivers 98% of GPT‑6.1 Sol’s performance at 40% of the price, and 96% of Anthropic’s Claude Opus 5.5 performance at 5% of the cost. The smaller, faster MAI Code 2 Flash is described as matching GPT‑6 Luna at 65% of the price and Claude Haiku 5.5 at 20% of the cost. For cybersecurity, MAI Cyber 2 Flash is said to beat GPT‑6 Astra on a security benchmark while operating at less than 2% of the cost per task.
In image generation, Microsoft states that MAI Image 2.7 ranks first in image editing. New transcription and real‑time voice models are also intended to challenge existing competitors in those domains. Microsoft notes that its performance comparisons rely on selected benchmarks and that some pricing advantages reflect promotional discounts, while still positioning the systems as cost‑efficient alternatives.
Specialized model strategy and cost focus
Mustafa Suleyman, Microsoft’s AI chief, describes a strategy of specializing models for particular tasks so they can offer comparable performance at significantly lower costs. He characterizes the approach as “cram more tokens and knowledge and expertise into the same amount of parameters,” with the trade‑off that a model optimized for everyday coding will be less capable at advanced math or scientific research. Suleyman also says developing models tailored to specific tasks is helping Microsoft control costs while remaining competitive with other leading AI developers.
He adds that the company’s approach is intended to improve both performance and affordability over time rather than hinge on a single breakthrough system. This positioning underscores a shift toward a portfolio of domain‑specific tools, including for software development and cybersecurity, instead of relying solely on general‑purpose models.
Integration, enterprise access and long‑term ambitions
MAI Code 2 will be available in GitHub Copilot, which could move some coding workloads from outside models onto Microsoft’s own infrastructure. Suleyman declined to specify how much GitHub Copilot traffic might shift, but he expects significant adoption. The cybersecurity‑focused model will initially roll out to a select set of customers as organizations contend with increasing software vulnerabilities uncovered with the help of AI tools.
For enterprise users, Microsoft is offering access to licensed model weights so customers can customize models with proprietary data while still deploying them within Azure. Suleyman says this setup is intended to “feel like open source” while providing greater security and control. Alongside the specialized MAI models, Microsoft is working on a frontier general‑purpose large language model and reports state‑of‑the‑art pre‑training and strong progress in reinforcement learning, though no release timeline has been provided. Suleyman describes his team’s long‑term aim as building “true superintelligence.”
Competitive landscape with OpenAI
The launches come as OpenAI introduces new systems GPT‑6.2 Astra and GPT‑6.2 Sol, with the latest Astra priced 20% below the previous version. This parallel round of releases highlights active competition on both capability and pricing among leading AI developers. Microsoft’s combination of specialized MAI models, integration into products such as GitHub Copilot, enterprise customization options and work on a frontier model positions it to challenge rivals across multiple application areas.
Key Takeaways
- 01Microsoft is deepening its in‑house MAI portfolio to reduce reliance on external model providers while broadening coverage from coding to cybersecurity and voice.
- 02Cost efficiency is central to the new lineup, with Microsoft emphasizing benchmarked parity or near‑parity to rival models at substantially lower listed prices.
- 03Embedding MAI Code 2 into GitHub Copilot and offering model weights for Azure‑based customization aim to lock AI workloads into Microsoft’s ecosystem.
- 04The company is pursuing both specialized task models and a frontier general‑purpose system, signaling a dual track of near‑term products and longer‑term ambitions.
References
- https://www.axios.com/2026/11/17/microsoft-ai-models-openai-anthropic-ignite#utm_source=yahoo_finance&utm_medium=partner&utm_campaign=subs-partner-yahoo-finance-AI?.tsrc=rss
- https://mezha.net/eng/news/mezha-16f9dab6ea0be270c87644f9_microsoft_targets_openai
- https://www.cnbctv18.com/technology/microsoft-takes-on-openai-and-anthropic-with-new-ai-models-for-coding-and-cybersecurity-20036327.htm
- https://cryptobriefing.com/microsoft-unveils-six-mai-models-ignite/