# Microsoft AI launches MAI-Image-2.5-Pro and MAI-Voice-2-Flash in public preview > Source: > Published: 2026-07-23 17:49:47+00:00 On July 23rd 2026 Microsoft AI announced the public preview of two purpose‑built generative models – MAI-Image-2.5-Pro and MAI-Voice-2-Flash – as part of its growing MAI family. The rollout follows a year‑long effort by the company's Superintelligence team to develop in‑house models trained on clean, enterprise‑grade data, a strategy Microsoft says lets it tailor performance, cost and latency to specific product needs. **MAI-Image-2.5-Pro** is billed as the highest‑fidelity image model in the MAI lineup. It supports “hero imagery, detailed editing, precise in‑image text rendering” and is priced at $5 per 1 M text input tokens, $8 per 1 M image input tokens, and $106 per 1 M image output tokens. The model’s claim on the blog post is that it delivers “best‑in‑class Arena ELO scores at a lower price” and can render text within images with a level of accuracy the company calls a “real breakthrough.” **MAI-Voice-2-Flash** focuses on speed and scale for voice applications. Microsoft says Flash is twice as fast as its predecessor MAI-Voice-2 and 32% cheaper, with a price of $15 per 1 M characters. The model retains the natural prosody of MAI-Voice-2 while reducing latency, making it suitable for high‑volume, low‑latency scenarios such as call‑center agents. Both models sit alongside existing production models, giving developers a “quality‑speed‑cost curve” to choose from. The blog notes that the models are already powering a range of Microsoft services: **Bing Image Creator** now runs end‑to‑end on MAI-Image-2.5, giving users finer control over generation and editing. **PowerPoint** uses the model for image‑to‑image capabilities, delivering up to an 84% reduction in GPU costs versus the earlier GPT‑Image‑2. **OneDrive** has made MAI-Image-2.5 the default for key image‑editing scenarios, reporting a 26% increase in save rates, a roughly 25% cut in P95 latency, and 2.5× greater efficiency under medium‑utilization workloads. **Dynamics 365 Contact Center** now runs MAI-Voice-2-Flash for voice agents, delivering expressive speech while cutting GPU costs by up to 89%. **Azure Voice Live** offers developers a scalable path to build high‑quality speech agents using the Flash model. The announcement also highlights a partnership with **Dragon Copilot**, a medical documentation solution used by about 170,000 providers. MAI-Transcribe-1.5, another model in the MAI suite, replaces the previous model in Dragon Copilot, delivering a 50% relative reduction in transcription and language‑identification errors across 58 languages. Rob Reilly, Global Chief Creative Officer at WPP, is quoted in the post: “MAI-Image-2.5-Pro is a strong leap forward for GenMedia tools. Beyond the impressive image quality, its ability to render text with this kind of accuracy is a real breakthrough. It also understands natural language edits, so creative iteration becomes faster and far more intuitive. Microsoft has firmly established itself among the leaders in generative AI.” Microsoft frames the launch as evidence that its in‑house model strategy is moving beyond leaderboard bragging to real‑world product impact. The company asserts that the models are “running in production, at scale, efficiently powering experiences for millions of users across a growing portfolio of Microsoft products.” The pricing structure suggests a clear commercial play: higher‑quality image generation remains relatively expensive per output token, while voice generation is positioned as a low‑cost, high‑throughput option for enterprise workloads. By offering multiple variants, Microsoft appears to be hedging against a one‑size‑fits‑all approach and signaling to developers that they can optimize spend without sacrificing performance. Developers can start experimenting with the models in the **MAI Playground** or via the **Foundry** platform, both linked from the Microsoft AI blog. The move deepens Microsoft’s AI moat against third‑party providers such as OpenAI and Anthropic, whose models continue to dominate many generative‑AI use cases. By keeping the core model pipeline internal, Microsoft can align updates directly with product roadmaps for Bing, Office, Azure and Dynamics, potentially tightening integration and pricing advantages. The public preview marks the latest step in Microsoft’s broader AI agenda, which includes embedding proprietary models across its consumer and enterprise services. As the company continues to replace externally sourced models with its own, the competitive landscape for generative AI services may shift toward tighter vertical specialization and cost‑based differentiation.