Microsoft's MAI Code 2 claims 96% of Claude Opus 5.5's power for 5% of the price
Microsoft's Ignite 2026 keynote brought two homegrown coding models and a shot across OpenAI's bow. The company says MAI Code 2 delivers 96% of Claude Opus 5.5's performance at 5% of the price, with Code 2 Flash going cheaper still. Both are landing as optional picks inside Copilot, Microsoft 365 and Fabric.
- Original source: zdnet.com.
- The source published it on Nov 17, 2026 at 10:42 PM.
- No second outlet has reported it yet. Treat it as a single-source claim.
Microsoft used its Ignite 2026 keynote to roll out two in-house frontier models built for what it calls “everyday” coding work. MAI Code 2, the headline act, delivers 96% of Claude Opus 5.5’s performance at 5% of Opus 5.5’s price, according to the company. The cheaper Code 2 Flash sits alongside GPT-6 Luna and Claude Haiku 5.5 in capability, again at a considerably lower cost.
There’s a catch on the sticker shock: these are usage-based API figures, not subscription math, so a $100-a-month plan doesn’t morph into a five-dollar one. What matters is the posture. Microsoft is now benchmarking its own engines against Anthropic and OpenAI, and is wiring them in as optional choices for GitHub Copilot, Microsoft 365 and Fabric. Satya Nadella’s slides leaned on a spread of foundation models meant to plug into Copilot and the rest of the stack.
The OpenAI subplot writes itself. Microsoft’s multi-billion-dollar bet on the lab dates to 2023, and plenty of its products are built on OpenAI intellectual property. Every routine AI request across that product line has meant token fees flowing back to frontier providers — or customers paying them instead. ZDNET notes there’s no direct evidence Microsoft is trying to build replacements for those providers, but the vertical integration is hard to miss.
A third model joined the party: MAI Work Flash, which Microsoft says was derived from Code 2 Flash. It generates Excel formulas, works with Word documents and instructions, tracks decisions and pulls information out of Teams. ZDNET’s David Gewirtz points out the intelligence these tasks demand isn’t enormous — as long as a model is good enough, it’s good enough.
Gewirtz also runs the savings as a hypothetical: a company paying $0.01 per AI request could be paying $0.001 instead, and a business with a daily inference cost of a million dollars could cut that to roughly $100,000. Across a year, he calculates, the switch would save about $325 million. Microsoft says its own reported figures already bake in a 50% introductory discount, which is why ZDNET measured against $0.001.
Microsoft is clear about the ceiling. These models aren’t meant for the hardest agentic and research workloads — think reformatting your PowerPoint, not cracking a decades-old mathematics problem. They did unusually well on visualization assignments, a snug fit for UI design, graphics and slide production. Gewirtz flags benchmark selection as a possible cherry-pick, though he suspects Microsoft isn’t padding the numbers much, since it benefits from the cheaper models more than its own customers do.
TopicsMicrosoft · Satya Nadella · MAI Code 2 · Code 2 Flash · MAI Work Flash · Claude Opus 5.5 · OpenAI · GitHub Copilot
Source: zdnet.com
Written by AI from verified reporting.
More stories
- Tech & AI blockchain.news 0 viewsNvidia Ties Up Kinsus, Samsung, Amkor to Scale AI Hardware Supply
- Tech & AI cnbc.com 0 viewsNvidia tops estimates; neoclouds rally as Huang offers CEO return
- Tech & AI x.com 0 viewsAnthropic's Sequoia and Lightspeed Backers Plan Billions More in IPO
- Tech & AI x.com 0 viewsNvidia Introduced Alibaba to Alice Huang Before 2022 Chip Curbs, Report Says

