Microsoft has launched MAI-Image-2-Efficient, a streamlined version of its advanced image generation model. This new iteration promises higher quality visuals at a significantly reduced cost and increased speed. Developed by Microsoft's MAI superintelligence team, the model offers four-times faster throughput on Nvidia H100 processors and demonstrates 22% faster raw performance compared to its predecessor. Microsoft claims it outperforms Google's Gemini 3.1 Flash with 40% lower latency.
MAI-Image-2-Efficient also boasts cost savings, with pricing set at $5 per million input tokens and $19.50 per million output tokens, a 41% reduction. This dual-pricing strategy allows for high-fidelity creative work or cost-effective volume production, making it ideal for tasks like UI mockups and marketing assets.
This development is part of Microsoft's broader strategy to reduce its dependence on OpenAI. The company has historically relied heavily on OpenAI's models but is now diversifying its AI capabilities. Recent reports indicate growing friction between the two companies, with OpenAI exploring partnerships with competitors like Amazon Web Services and diversifying its cloud infrastructure. Microsoft has even officially listed OpenAI as a competitor.
Developing proprietary models like MAI-Image-2-Efficient allows Microsoft to retain internal costs and accelerate its agentic AI strategy. These models are now default options for services like Copilot, replacing OpenAI's DALL-E. This move is crucial for Microsoft's ambition to create AI that can execute complex, multi-step tasks and workflows for users, where speed and cost-efficiency are paramount for large-scale iteration.