Home
Microsoft Debuts Self-Developed Models MAI-Image-2.5-Pro and MAI-Voice-2-Flash, Now Integrated into Bing and PowerPoint
On July 23, Microsoft announced the public preview of two proprietary AI models: MAI-Image-2.5-Pro and MAI-Voice-2-Flash. These models are designed for high-quality image generation and high-concurrency voice interactions. Microsoft emphasized that the training data is cleaned, traceable, and not derived from third-party model distillation. Built from the ground up, they directly serve Microsoft product users.

Highest-precision image model reduces GPU costs by up to 84%
MAI-Image-2.5-Pro is Microsoft's most precise image model, targeting demanding use cases like hero image generation, detailed editing, and in-image text rendering, while supporting natural language editing instructions. It has become the default image generation model in Bing Image Creator and is used for image-to-image editing in PowerPoint. Compared to GPT-Image-2, it cuts GPU costs by up to 84%. After deployment in OneDrive, scenario success rates rose by 26%, P95 latency dropped by about 25%, and efficiency in medium-load production environments improved by 2.5 times.
Pricing is $5 per million tokens for text input and $106 per million tokens for image output.
Voice model doubles speed, serving customers like T-Mobile

MAI-Voice-2-Flash is optimized for high-frequency voice applications, offering roughly twice the speed of the previous generation and a 32% cost reduction, making it ideal for fast-response scenarios such as contact centers and voice assistants. Integrated into Dynamics 365 Contact Center, it serves customers like T-Mobile and EasyJet, with GPU cost reductions of up to 89%. Microsoft has also embedded it into the Azure Voice Live service, enabling developers to build voice-to-voice interactive agents.
Additionally, the MAI-Transcribe-1.5 model from the same series has been applied to the medical voice solution Dragon Copilot, used by 170,000 healthcare providers. It processed 28 million patient consultations last quarter, supports 58 languages, and reduces transcription error rates by roughly 50% for most languages. Microsoft revealed that the next-generation GB200 computing cluster is already operational, and the MAI series will continue to expand.
Related article
Mozilla Partners with Mistral to Bring AI Models to Firefox Smart Window
Mozilla has partnered with French AI startup Mistral to integrate the Mistral model into Firefox, enhancing the experimental "Firefox Smart Window" browsing experience.Designed for complex, multi-tab tasks, the Smart Window helps users analyze search
OpenAI reportedly exploring smartphone with AI agents replacing traditional apps
Speculation regarding OpenAI’s hardware strategy has intensified, with recent reports pointing toward the development of a dedicated smartphone. Industry analyst Ming-Chi Kuo indicates that the AI firm is collaborating with chipmakers MediaTek and Qu
Google Tests Gemini Daily Brief Text Customization for Precise Fine-Tuning
Google is rolling out a direct prompt customization feature for Gemini Daily Brief in the Google App 17.43.11 beta, enabling users to tailor their morning AI summaries via text, voice commands, or preset preferences. This update replaces the previous
Related Special Topic Recommendations
Comments (0)
0/500
On July 23, Microsoft announced the public preview of two proprietary AI models: MAI-Image-2.5-Pro and MAI-Voice-2-Flash. These models are designed for high-quality image generation and high-concurrency voice interactions. Microsoft emphasized that the training data is cleaned, traceable, and not derived from third-party model distillation. Built from the ground up, they directly serve Microsoft product users.

Highest-precision image model reduces GPU costs by up to 84%
MAI-Image-2.5-Pro is Microsoft's most precise image model, targeting demanding use cases like hero image generation, detailed editing, and in-image text rendering, while supporting natural language editing instructions. It has become the default image generation model in Bing Image Creator and is used for image-to-image editing in PowerPoint. Compared to GPT-Image-2, it cuts GPU costs by up to 84%. After deployment in OneDrive, scenario success rates rose by 26%, P95 latency dropped by about 25%, and efficiency in medium-load production environments improved by 2.5 times.
Pricing is $5 per million tokens for text input and $106 per million tokens for image output.
Voice model doubles speed, serving customers like T-Mobile

MAI-Voice-2-Flash is optimized for high-frequency voice applications, offering roughly twice the speed of the previous generation and a 32% cost reduction, making it ideal for fast-response scenarios such as contact centers and voice assistants. Integrated into Dynamics 365 Contact Center, it serves customers like T-Mobile and EasyJet, with GPU cost reductions of up to 89%. Microsoft has also embedded it into the Azure Voice Live service, enabling developers to build voice-to-voice interactive agents.
Additionally, the MAI-Transcribe-1.5 model from the same series has been applied to the medical voice solution Dragon Copilot, used by 170,000 healthcare providers. It processed 28 million patient consultations last quarter, supports 58 languages, and reduces transcription error rates by roughly 50% for most languages. Microsoft revealed that the next-generation GB200 computing cluster is already operational, and the MAI series will continue to expand.
Mozilla Partners with Mistral to Bring AI Models to Firefox Smart Window
Mozilla has partnered with French AI startup Mistral to integrate the Mistral model into Firefox, enhancing the experimental "Firefox Smart Window" browsing experience.Designed for complex, multi-tab tasks, the Smart Window helps users analyze search
OpenAI reportedly exploring smartphone with AI agents replacing traditional apps
Speculation regarding OpenAI’s hardware strategy has intensified, with recent reports pointing toward the development of a dedicated smartphone. Industry analyst Ming-Chi Kuo indicates that the AI firm is collaborating with chipmakers MediaTek and Qu
Google Tests Gemini Daily Brief Text Customization for Precise Fine-Tuning
Google is rolling out a direct prompt customization feature for Gemini Daily Brief in the Google App 17.43.11 beta, enabling users to tailor their morning AI summaries via text, voice commands, or preset preferences. This update replaces the previous











