Microsoft has introduced MAI Realtime 1, its first speech-to-speech AI model designed for natural, real-time conversations, with pricing starting at just $0.01 per audio minute. The launch is part of Microsoft’s broader release of seven new AI models, expanding its in-house AI portfolio across voice, coding, cybersecurity, image generation, transcription, and decision-making.

MAI Realtime 1 is designed to power conversational AI agents that can listen, respond naturally, handle interruptions, understand background noise, and perform actions through connected enterprise tools. The launch highlights Microsoft’s strategy of building powerful AI systems that businesses can deploy at scale without the high costs traditionally associated with advanced AI models.

What Is Microsoft MAI Realtime 1?

MAI Realtime 1 is Microsoft’s first dedicated speech-to-speech AI model. Instead of treating speech recognition, language processing, and voice generation as separate steps, it is designed to support a more continuous and natural conversational experience.

The model is intended to make voice interactions feel less robotic by responding quickly, understanding conversational context, and handling interruptions without forcing users to wait for rigid turn-taking.

Microsoft says the model is designed around several key capabilities:

  • Natural conversation: Handles interruptions, quick acknowledgments, and understands when a speaker has finished talking.
  • Adaptive intelligence: Responds quickly to simple requests while accessing deeper reasoning when more complex tasks require it.
  • Audio understanding: Can interpret speech in environments with background noise or multiple voices.
  • Real-world actions: Connects with enterprise tools and systems to complete tasks rather than merely answer questions.
  • Customizable voice behavior: Developers can adjust its tone, style, personality, and responses for different applications.

These capabilities could make voice AI more practical for customer support, business assistants, automated receptionists, and other conversational applications.

MAI Realtime 1 Pricing Starts at $0.01 Per Minute

One of the most significant parts of Microsoft’s announcement is its pricing strategy. MAI Realtime 1 starts at $0.01 per audio minute for its base speech-to-speech usage.

However, reasoning costs are billed separately, meaning businesses using more complex reasoning capabilities will have additional expenses.

Pricing componentCost
Base speech-to-speech usage$0.01 per audio minute
Reasoning input$1 per million tokens
Reasoning output$5 per million tokens

The pricing structure is designed to make continuous voice interactions more economical for businesses operating AI agents at scale.

For example, a voice AI system handling 10,000 minutes of audio would have a base speech-to-speech cost of approximately $100, excluding additional reasoning charges.

The lower pricing could be particularly important for businesses deploying voice agents for customer support, sales assistance, appointment booking, and automated service operations, where thousands of conversations may take place every day.

Microsoft Claims MAI Realtime 1 Outperforms GPT Realtime 2.1

Microsoft claims MAI Realtime 1 performs better than GPT Realtime 2.1 across quality, speed, and pricing in its evaluations.

The company says it evaluated the model using public benchmarks, including Big Bench Audio for speech reasoning and Tau-Voice for agentic task completion, alongside internal Dynamics 365 Contact Center evaluations.

Microsoft says MAI Realtime 1 sits on the performance-price frontier across these evaluations, suggesting that the model offers a competitive combination of capability and cost.

However, these comparisons are based on Microsoft’s own testing configurations and workloads. Independent testing and broader real-world usage will be important for determining how the model performs against competing voice AI systems across different applications.

OpenAI Realtime API Compatibility Makes Migration Easier

Microsoft has also made MAI Realtime 1 compatible with the OpenAI Realtime API.

This compatibility could help developers who already have voice applications built around the OpenAI Realtime API experiment with Microsoft’s model without rebuilding their entire infrastructure from scratch.

Developers may be able to adapt their existing voice application architecture, change model configurations, and evaluate the new system while keeping much of their existing integration workflow.

This could lower the barrier for companies considering alternatives to their current speech-to-speech AI provider, especially if pricing, latency, or performance is a concern.

Microsoft Targets Customer Service and Enterprise Voice Agents

Microsoft is positioning MAI Realtime 1 primarily for enterprise applications rather than just consumer voice assistants.

The company says the model is being tested by businesses including Agoda and Reservations.ai. Microsoft is also planning integrations with its enterprise products and services.

Two applications specifically highlighted by Microsoft include:

Dynamics 365 Contact Center

MAI Realtime 1 is being brought to Dynamics 365 Contact Center to support customer-service applications. AI-powered voice agents could help businesses handle customer queries, provide information, assist with service requests, and automate repetitive support interactions.

Dragon Copilot

Microsoft is also bringing the model to Dragon Copilot for simulated patient conversations. The application is designed to help nurses practice clinical interactions using AI-generated patient scenarios.

These integrations illustrate Microsoft’s broader approach of embedding AI models directly into existing enterprise software rather than offering them only as standalone products.

Microsoft’s Seven New MAI Models

MAI Realtime 1 is only one part of Microsoft’s latest AI launch. The company introduced seven models targeting different workloads and business requirements.

AI modelPrimary purpose
MAI Code 2Advanced coding and software engineering
MAI Code 2 FlashFaster, lower-cost coding tasks
MAI Cyber 2 FlashCybersecurity and threat-related workloads
MAI Image 2.7Image generation and editing
MAI Realtime 1Real-time speech-to-speech conversations
MAI Transcribe 2Multilingual speech transcription
Decision 1.1Model routing and decision-making

The lineup reflects Microsoft’s attempt to develop specialized AI systems for individual tasks, rather than relying exclusively on one general-purpose model for every workload.

Microsoft is emphasizing price-performance, with several models positioned as lower-cost alternatives to competing systems while maintaining strong benchmark results.

Why Microsoft Is Building Its Own AI Models

The launch represents a broader strategic shift for Microsoft as AI infrastructure costs rise and businesses increasingly deploy autonomous agents.

Microsoft has invested heavily in AI infrastructure, cloud computing, data centers, and partnerships, but developing proprietary models gives the company more control over model performance, deployment costs, and product integration.

In-house models can also help Microsoft differentiate its cloud and software offerings. Instead of depending entirely on external AI providers, Microsoft can integrate its own models into Azure, GitHub Copilot, Dynamics 365, Microsoft healthcare products, and other enterprise services.

For customers, the strategy could create more options based on the workload. Businesses may choose a premium model for complex tasks and a cheaper specialized model for routine interactions.

The Bigger Picture

The launch of MAI Realtime 1 demonstrates how the AI industry is moving beyond text-based assistants toward systems capable of handling voice conversations and real-world tasks.

Voice AI is becoming increasingly important for customer service, enterprise automation, healthcare training, sales workflows, and digital assistants. However, businesses need more than impressive demonstrations. They need predictable costs, low latency, reliable responses, integration with existing systems, and the ability to operate at scale.

At $0.01 per audio minute, Microsoft’s pricing could encourage more companies to experiment with conversational AI agents. But total operating costs will still depend on reasoning usage, infrastructure, integrations, and the complexity of the tasks being handled.

Looking Ahead

MAI Realtime 1 will initially be available through Microsoft’s MAI Labs experience and a limited-access preview through Microsoft Foundry. Microsoft says preview customers will not be billed until public preview begins. Wider availability, independent performance testing, and enterprise deployments will determine how quickly businesses adopt the model.

Microsoft’s seven-model launch also signals that the company wants to compete directly across multiple AI categories, not just through its cloud platform and partnerships. If these models deliver competitive performance at substantially lower costs, Microsoft could strengthen its position in the growing market for enterprise AI agents, voice automation, and specialized AI software.

Get the day’s top stories in your inbox

One concise email. No spam, unsubscribe anytime.