Microsoft has unveiled MAI-Image-2-Efficient, an entry-level version of its AI image generation model.

On April 15, online media outlet Gigazine reported that the model is based on the same architecture as MAI-Image-2 while touting lower costs and faster generation speed.

The new model is an efficiency-improved version of MAI-Image-2, which Microsoft introduced on March 19. When generating 1024 by 1024 images on Nvidia H100, it can process up to 22 percent faster than the existing model. Microsoft said it cut costs by 41 percent while maintaining flagship-level quality.

The performance differentiation also extends to the character of the generated output. Microsoft explained that the new model tends to draw sharp, clear lines, making it suitable for illustrations, animation and attention-grabbing photorealistic image generation. By contrast, it said MAI-Image-2 highlights smooth, detailed depiction, and recommended using the two models separately depending on the purpose.

Microsoft also stressed speed in comparisons with rival models. In data comparing 1-megapixel image generation times, it presented MAI-Image-2-Efficient as capable of faster generation processing than Gemini 3 Pro Image, Gemini 3.1 Flash Image and GPT-Image-1.5-High. By emphasizing both speed and cost, the move targets corporate demand to attach image-generation AI to work tools or services.

The model is currently available on Microsoft Foundry and MAI Playground. Under Foundry pricing, the API costs $5 per 1 million tokens for text input and $19.5 per 1 million tokens for image output. Microsoft also released a model card so specifications and usage conditions can be checked.