Copilot goes cheap as Microsoft phases out OpenAI and Anthropic models to cut costs

2026-07-07 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Novice, quick

Summary

Microsoft is phasing out OpenAI and Anthropic models in several Copilot products, including Excel and Outlook, replacing them with its own in-house MAI models. This strategic shift, reported by Bloomberg on July 7, 2026, aims to significantly cut costs, as acknowledged by Microsoft's head of AI, Mustafa Suleyman. While MAI models currently process tens of thousands of requests weekly in Excel and Outlook and are available in GitHub Copilot, they represent a small fraction of total usage. Microsoft's MAI-Thinking 1, a reasoning model unveiled at Build conference, was claimed to match Sonnet 4.6 and Opus 4.6 in coding, but benchmarks showed it trailing, performing closer to Deepseek V3.2. This transition could mean customers pay the same price for potentially less capable AI. Microsoft CEO Satya Nadella also suggested a move towards usage-based pricing, possibly offering MAI models as default and third-party models as premium add-ons. Furthermore, despite claims of commercially licensed data, MAI models reportedly use Common Crawl, a dataset with legally unsettled status for AI training.

Key takeaway

For AI Product Managers evaluating Copilot subscriptions, be aware that Microsoft's shift to in-house MAI models aims to cut their costs, not necessarily improve your experience. You should scrutinize model performance claims and prepare for potential usage-based pricing changes. Evaluate if default MAI models meet your needs, or budget for premium third-party add-ons if higher capabilities are critical.

Key insights

Microsoft is replacing third-party AI models with its own MAI models in Copilot to reduce costs, potentially impacting customer value and data claims.

Principles

• Cost reduction drives AI model strategy.
• In-house models reduce third-party reliance.
• Benchmarks reveal true model capabilities.

In practice

• Evaluate AI model performance independently.
• Scrutinize vendor claims on training data.
• Prepare for usage-based AI pricing shifts.

Topics

• Microsoft Copilot
• MAI Models
• AI Cost Reduction
• Model Benchmarking
• Usage-Based Pricing
• AI Training Data

Best for: Investor, CTO, VP of Engineering/Data, Director of AI/ML, AI Product Manager, Tech Journalist

Related on AIssential

• We Committed Fraud with OpenAI's New Image Model (and Called Mum) - EP99.38 — AI image generation has reached a level of realism enabling highly convincing forgeries, posing significant societal and economic challenges.
• Open-Source AI Gains Ground as Rising Costs Push Shift to Smaller Models — Open-weight AI models offer substantial cost savings and near-parity performance compared to closed models, yet remain significantly underutilized.
• Copilot just 9x'd Sonnet and 27x'd Opus and teams have no idea — AI model providers are ending subsidized pricing, shifting to usage-based billing reflecting true compute costs.
• LAI #131: A Tool Call Can Succeed and Still Be the Wrong Tool — Microsoft's MAI models prioritize human-generated data and transparent lineage over synthetic data for foundational capabilities.
• How AI Customers Are Lowering Their Anthropic and OpenAI Bills — AI customers are optimizing costs by switching to cheaper models from current or alternative providers.
• Microsoft Mulls Using DeepSeek for Copilot Cowork — Microsoft considers DeepSeek V4 or other open-source models for Copilot Cowork to reduce AI assistant operational costs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.