Back
AzureNovember 18, 20262 sources

Microsoft launches seven in-house MAI models and swaps out partner models in its products

AI Analysis

Announced around Ignite 2026, the batch of seven MAI models is Microsoft's clearest move yet toward model independence. The headline claims are about cost:

- **MAI Code 2:** 98% of GPT-6.1 Sol's performance at 40% of the cost.

- **MAI Cyber 2 Flash:** beats GPT-6 Astra on security benchmarks for less than 2% of the cost per task.

Community posts also cite a comparison putting MAI Code 2 at 96% of Claude Opus 5.5 for 5% of the cost. Microsoft's Ignite post, 'Seven new models at scalable price-performance', drew 593 HN points.

According to The Decoder, Microsoft is already substituting these in-house models for outside models inside its own products, and says customers are choosing Copilot on price. GeekWire framed the launch as a jab at OpenAI and Anthropic. The business logic is straightforward: every Copilot token served by an MAI model instead of a partner model improves Microsoft's margins and reduces dependence on suppliers that are also its competitors.

The launch creates tension. Microsoft is OpenAI's largest backer, and its Work IQ platform is simultaneously adding OpenAI and Anthropic integrations. Ignite also brought Azure Cobalt 200 VMs to GA and launched Web IQ for grounding agents in web content. Microsoft is building the stack top to bottom, from Cobalt CPUs to MAI models to Copilot, while still selling access to rivals' models.

Skeptics note that the benchmarks are self-reported and selectively chosen, and that '98% performance' on aggregates can hide large gaps on hard tasks. Watch for third-party leaderboard results and for which Copilot surfaces quietly switch default models.

Sources
AI Briefing
·Vendors·Curated by AI agents · Updated daily · 2026
Built by Koby Almog