Innovation & Tech

Microsoft Fires at Anthropic and OpenAI With Coding and Security AI

Microsoft launched MAI Code 2 and a cybersecurity AI model at Ignite, claiming near-rival performance at a fraction of the cost — with no independent benchmarks yet.

By Nathan Brooks

3 min read

Microsoft unveils coding and cybersecurity AI models at Ignite, targeting Anthropic and OpenAI
Microsoft unveils coding and cybersecurity AI models at Ignite, targeting Anthropic and OpenAIPexels / pixabay

What's News

  • Microsoft launched MAI Code 2 and a cybersecurity AI model at Ignite in San Francisco on Tuesday
  • Microsoft expects over $175 billion in capital expenditure this calendar year and plans to deploy more than a million Maia 300 chips next year
  • MAI-Base-2 is a 2.8 trillion-parameter MoS model claiming up to 12.3x better compute efficiency than DeepSeek-V4-Pro-Base
  • Anthropic's Claude Fable 5.2 and Mythos 5.2 launched November 9 at $8/$40 per million input/output tokens, with up to 30% faster outputs
  • No independent benchmarks, pricing or release dates were provided for the new Microsoft models

Microsoft used its Ignite conference in San Francisco on Tuesday to launch specialized AI models for software development and cybersecurity, taking direct aim at the two strongest players in those niches: Anthropic and OpenAI.

The centerpiece is MAI Code 2, a programming model that Microsoft says "almost matches the performance of rival AI models at a fraction of the cost." The company also unveiled a cybersecurity-specific model headed for a limited group of initial customers, plus new image-generation and transcription models.

For practitioners, the pitch is simple: domain-specific models tuned for one job, priced to undercut general-purpose rivals. AI chief Mustafa Suleyman told Reuters the approach helps "keep costs low while matching rivals and in some cases topping model benchmarks."

"This isn't just one model. It's not a flash in the pan," Suleyman said. "This is a system, a hill-climbing machine that we have built, which co-optimizes performance and price."

What do the claims actually rest on?

Not much public evidence, so far. The Reuters report on the announcement cited no benchmark data, no evaluation methodology, and no names for the "rival AI models" Microsoft compared itself against. The near-parity claim comes solely from the company.

Microsoft also stayed silent on release dates, pricing, and access terms for MAI Code 2. The cybersecurity model gets a limited customer rollout, which Microsoft justified by citing "a surge in software vulnerabilities that AI tools are increasingly bringing to light." Which customers, on what terms, with what capabilities — all unspecified.

The competitive backdrop

The launch lands weeks after Anthropic's November 9 release of Claude Fable 5.2 and Claude Mythos 5.2. Fable targets coding, reasoning and agent tasks; Mythos is restricted to approved cybersecurity and life-sciences research organizations. Anthropic claimed the new models generate outputs up to 30% faster than their 5.1 predecessors, priced at $8 per million input tokens and $40 per million output tokens — below earlier Fable and Mythos releases.

OpenAI, for its part, has widened distribution of frontier models including GPT-6 Astra through Google Cloud and Amazon Web Services since its exclusive Azure arrangement ended in April, per a Reuters report on October 21. Microsoft's MAI family is part of a stated strategy to reduce reliance on external providers — a notable shift given that history.

Alignment moves up the priority list

Suleyman pointed to a recent security episode — described by Reuters as "OpenAI agents' hacking of the Hugging Face platform" — as the catalyst for pairing safeguards with performance. He called AI alignment, meaning preventing AI from taking harmful actions and ensuring it behaves as intended, his top priority.

Reuters reported no independent detail on the incident, no comment from OpenAI on Suleyman's characterization, and no further specifics on its scope.

The hardware under the claims

The model push sits on a heavy infrastructure bet. According to Investing.com, Microsoft has finished pre-training MAI-Base-2, a 2.8 trillion-parameter mixture-of-experts model with 80 billion active parameters, claiming 4.2 to 12.3 times better compute efficiency than DeepSeek-V4-Pro-Base across four evaluations.

A Reuters report from October 12 said Microsoft plans to deploy more than a million of its custom Maia 300 chips next year, targeting "multi-gigawatt deployments." The same report put expected capital expenditure above $175 billion for this calendar year. Those numbers are what make the claimed cost advantages of smaller, specialized models plausible — though Microsoft hasn't disclosed how specific model economics map to that hardware spend.

What's still unknown

  • No independent benchmarks exist for MAI Code 2 or the cybersecurity model.
  • Pricing, availability timelines and access terms remain undisclosed for both.
  • Microsoft never named the rival models it benchmarked against.

The message to the market is clear: Microsoft intends to compete head-on at the application layer where Anthropic and OpenAI earn their keep. Whether the domain-specific bet delivers durable cost and performance wins now depends on execution — and on evidence Microsoft hasn't yet published.

Source: epyhia.com

Share this article:

More from Nathan Brooks

Nathan Brooks

Show full bio

News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.

2,349 articles

Related articles

« Previous articleNext article »