Skip to content
Bramble

Technology

Microsoft unveils MAI Code 2 and cybersecurity AI models at Ignite

At its Ignite conference on 17 November, Microsoft announced a suite of in-house AI models, including the coding model MAI Code 2, a cybersecurity model for a limited group of customers, and image and transcription models. Microsoft says the coding model nearly matches rivals' at a fraction of the cost; the benchmarks are Microsoft's own.

2 outlets · 0L · 2C · 0R First reported Account updated
Image: CNA
Image: ZDNET

1 / 2

The story, neutrally told

Centre · 1Microsoft on Tuesday 17 November announced a suite of AI models with coding and cybersecurity capabilities, as it competes with rivals such as Anthropic and OpenAI. Centre · 1At the Ignite conference in San Francisco, the company unveiled a programming model called MAI Code 2, which it says almost matches the performance of rival AI models at a fraction of the cost. Centre · 1According to ZDNET, Microsoft claims MAI Code 2 delivers 96% of the performance of Anthropic's Claude Opus 5.5 at 5% of its price, and is comparable to OpenAI's GPT-6.1 Sol.

Centre · 1A second, smaller model, Code 2 Flash, is described as comparable to the lower-end GPT-6 Luna and Claude Haiku 5.5, also at considerably lower cost. Centre · 1Microsoft also announced MAI Work Flash, derived from Code 2 Flash, aimed at tasks such as generating Excel formulas, working with Word documents and extracting information from Teams. Centre · 1The company also presented a cyber-specific model that it plans to offer to a limited group of customers, to help address a surge in software vulnerabilities that AI tools are increasingly bringing to light, and released new image-generation and transcription models.

Centre · 1Microsoft AI chief Mustafa Suleyman told Reuters that building technology specific to different domains helps keep costs low while matching rivals and in some cases topping model benchmarks, calling it "a system, a hill-climbing machine" that co-optimizes performance and price. Centre · 1Suleyman said the recent cybersecurity reckoning, illustrated by OpenAI agents' hacking of the Hugging Face platform, led him to focus on safeguards as well as performance, and that alignment, preventing AI from taking harmful actions, was his top priority. Centre · 1ZDNET cautions that all benchmarks are provided by Microsoft and may differ under independent testing, that the pricing figures reflect a 50% introductory discount, and that the savings apply to usage-based API pricing, not consumer subscriptions.

Centre · 1Microsoft said the models are not designed for the most challenging agentic and research workloads, and is integrating them as optional choices in GitHub Copilot, Microsoft 365 and Fabric. Centre · 1ZDNET's author argues the models could make Microsoft substantially less dependent on outside model providers such as OpenAI, in which it has invested billions since 2023, and illustrates with a hypothetical saving of $325 million a year; the article adds there is no direct evidence Microsoft is trying to replace the frontier providers' models.

Every sentence links to the reporting it rests on. The pill in front of each says where its sources sit: Left, Centre or Right when one side supplies at least half of them, Mixed when they are evenly split. The number is how many outlets it cites.

Left0 outlets

No left outlet in our sources has covered this story yet.

Centre2 outlets

Framing
CNA reports a news-style account of the product launch and Suleyman's comments on cost and safety; ZDNET offers analysis of what cheaper in-house models mean for Microsoft's dependence on OpenAI and Anthropic.
Emphasis
CNA stresses the cybersecurity model and alignment priorities; ZDNET stresses the 96%-at-5%-of-cost claim, cost savings and caveats about benchmarks.
Leaves out or plays down
CNA gives no specific benchmark or price figures or caveats on Microsoft's claims; ZDNET does not mention the cybersecurity model or Suleyman's safety remarks.
Charged language
“races to challenge”“a hill-climbing machine”“lies, damned lies, and benchmark results”
For example
“as it races to challenge rivals such as Anthropic and OpenAI.” — CNA
“But there’s a bigger story: what these models might mean for Microsoft’s AI costs and its dependence on OpenAI.” — ZDNET

Right0 outlets

No right outlet in our sources has covered this story yet.

Ask about this story

Loading…