Microsoft has introduced MAI-Thinking 1, its first reasoning model, designed to deliver performance comparable to Anthropic's Claude Opus 4.6 for programming tasks while costing significantly less. This marks a pivotal moment in the AI landscape, where major tech companies are increasingly building their own models rather than relying on external vendors like OpenAI and Anthropic.

Why Is Microsoft Building Its Own Reasoning Models?

Microsoft's strategy reflects a clear financial calculation. The company processes enormous volumes of AI tokens through services like Copilot, which handle millions of prompts weekly. While Microsoft currently benefits from favorable pricing terms with OpenAI due to their long-standing partnership, the company wants to insulate itself from future price increases by external model providers. According to Bloomberg reporting, tens of thousands of AI requests are now being processed weekly by Microsoft's own MAI models within applications such as Excel and Outlook.

"Anthropic is extremely expensive, and many organizations are urgently seeking alternatives," stated Mustafa Suleyman, who leads the development of Microsoft's AI models.

Mustafa Suleyman, AI Model Development Lead at Microsoft

Suleyman further noted that the ultimate goal is to completely eliminate the costs associated with using Anthropic models. This aggressive posture underscores how seriously Microsoft takes the economics of AI deployment at scale.

How Does Microsoft's Reasoning Model Compare to OpenAI's o-Series?

OpenAI's o-series models, including o4-mini, o3, and o3-pro, are specifically built for deep reasoning tasks in mathematics, science, code, and planning. These models spend hidden reasoning tokens on hard problems before generating answers. In mid-2026, OpenAI's pricing for the o-series reflects the computational cost of this approach.

Microsoft's MAI-Thinking 1 enters this competitive space by claiming performance parity with an earlier generation of Claude Opus for programming tasks, while maintaining a cost advantage. The model is part of a broader MAI family that Microsoft unveiled during its Build developer conference in June, which includes specialized models for image generation, transcription, speech recognition, and programming tasks.

Steps to Understand OpenAI's Current Model Lineup and Pricing

• General-Purpose Models: OpenAI's GPT-5 series serves as the workhorse for most applications, with GPT-5.4 priced at $2.50 per million input tokens and $15 per million output tokens, making it the default choice for most teams.
• Reasoning Models: The o-series includes o4-mini at $1.10 input and $4.40 output per million tokens, o3 at $2 input and $8 output after a July 2026 price cut, and o3-pro at $20 input and $80 output for the most demanding reasoning tasks.
• Specialized Models: OpenAI offers purpose-built models including GPT-5.3-Codex for agentic coding at $1.75 input and $14 output per million tokens, gpt-realtime-2.1 for live voice interactions, and gpt-image-2 for image generation and editing.

The o-series models operate with a 200,000-token context window, smaller than the GPT-5 line's approximately one million-token window. This trade-off reflects the different computational demands of reasoning versus general-purpose tasks.

Output tokens represent the real cost driver across OpenAI's lineup. GPT-5 nano output costs $0.40 per million tokens, while GPT-5.5 output reaches $30 per million tokens, representing a 75-fold difference. The pro tiers push even higher at $180 per million output tokens.

What Does This Mean for Enterprise AI Strategy?

Microsoft's expansion of in-house AI capacity signals a broader industry trend. For now, the use of Microsoft's own models accounts for only a small portion of the company's total AI usage, but the trajectory is clear. Microsoft is gradually expanding the use of its own AI technology across its software portfolio, with plans to roll out internally developed transcription models to Teams and other products in the coming months.

The company is also deploying its own MAI models within GitHub Copilot, further reducing dependence on external vendors. This diversification strategy allows Microsoft to maintain control over costs while ensuring consistent performance across its ecosystem of applications and services.

For organizations evaluating AI models, the landscape now includes not just OpenAI and Anthropic, but also Microsoft's growing portfolio of reasoning and specialized models. Engineers and product teams must now consider cost-per-task, reasoning requirements, and context window needs when selecting between general-purpose models like GPT-5.4, reasoning specialists like o3, or Microsoft's emerging alternatives.

The competitive pressure is intensifying. As Microsoft demonstrates that reasoning capabilities can be achieved at lower cost, other enterprises may follow suit, building or licensing alternative models rather than accepting premium pricing from established vendors. This shift could reshape how organizations budget for AI infrastructure and which vendors they prioritize in their technology stacks.