Skip to content

Microsoft Routes Queries in Office Apps to Its Own Models

Bottom line: Microsoft is diverting token-intensive queries to less powerful internal models to reduce costs for external APIs.

Microsoft is gradually shifting queries from Excel and Outlook away from OpenAI and Anthropic to internal models from the MAI family. The background is economic pressure from high token costs at intensive usage levels.

Microsoft is currently redirecting several tens of thousands of user prompts per week in Excel and Outlook from OpenAI and Anthropic models to its own Microsoft AI (MAI) model family. This is reported by Bloomberg citing internal information. The scope has so far been small, but demonstrates Microsoft’s strategy to reduce dependence on third-party providers.

Cost pressure drives the transition: While flat-rate subscriptions are profitable for average users, internal analyses show problematic scenarios for heavy users. A single user with a 200-dollar subscription can incur token costs of up to 14,000 dollars per month. The mobility company Uber illustrates the extent: it consumed its annual AI budget for 2026 in four months because developers systematically deployed automated tools like Claude Code. Mustafa Suleyman, head of Microsoft’s AI division, said in June 2026: “We pay a lot of money to Anthropic – our goal is therefore to reduce these costs and ultimately to eliminate them.”

The performance gap persists: Microsoft presented seven new models at the Build conference in June 2026, including MAI-Thinking 1 for logical reasoning. Microsoft claims competitiveness with Claude Sonnet and Claude Opus for programming tasks. Independent benchmarks from specialist publication The Decoder contradict this: MAI models lag significantly behind leading Western models in performance and are more similar to open-source solutions like DeepSeek.

Cost pressure is accelerating migration to Chinese providers. While premium Western models cost four dollars or more per million tokens, Chinese service providers offer comparable core capabilities from 18 cents per million tokens. For CTOs this means: price competition is intensifying, proprietary models are currently inadequate for demanding tasks, and alternatives to established providers are growing.


Source: www.it-daily.net · Published 11 July 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.

Share on: