Microsoft launches MAI-Image-2-Efficient with 41% cost cut

Generating AI images for cool product mockups is one thing. Doing it at scale without burning a hole in your engineering budget is another.

Microsoft rolled out MAI-Image-2-Efficient, a leaner text-to-image model engineered specifically to cut inference bills while keeping flagship-level quality. It is available right now in Microsoft Foundry and MAI Playground with no waitlist, and is slated to land in Copilot, Bing, and PowerPoint soon.

Cheaper, faster, and built for bulk work

The most eye-catching part is the price sheet. Microsoft priced MAI-Image-2-Efficient at $5 per 1M text input tokens and $19.50 per 1M image output tokens, marking a roughly 41% discount compared to its predecessor.

It is also 22% faster and boasts 4x the GPU throughput efficiency of MAI-Image-2 when benchmarked on NVIDIA H100 hardware at 1024×1024. In median P50 render speed tests against rivals like Gemini 3.1 Flash Image, Gemini 3 Pro Image, and GPT-Image-1.5-High, Microsoft clocked it at roughly 40% faster on average.

Shutterstock is already putting it through its paces. Vanessa Salvo, Principal Product Manager at Shutterstock, noted that the model shows strong gains in prompt fidelity and usability, which matters most when teams transition from experimental hacking to high-volume production pipelines.

Microsoft is pitching this as a clear two-tier setup. The Efficient version takes care of high-volume runs like product shots, UI mocks, social media assets, and short in-image labels. The flagship MAI-Image-2 remains the scalpel for high-detail portraits, intricate artistic styles, and heavy in-image text.

Why it matters

For builders and creators, this significantly lowers the barrier to running real-time generative image features inside customer-facing apps. If you previously held back on batch generation or live image editing tools because of per-image API costs, a 41% cut combined with faster turnaround makes unit economics actually work.

Across the industry, the spotlight is shifting away from benchmark vanity metrics toward cost per successful render. By aggressively driving down inference expenses and funneling this model directly into office workflows like PowerPoint, Microsoft is putting serious pressure on Google and OpenAI to trim their own image generation fees.

We think Microsoft's two-model strategy is remarkably pragmatic. Instead of trying to force one giant, expensive model to do everything, delivering a specialized workhorse allows businesses to scale without getting crushed by GPU invoices. The only catch looks to be expectations management: users will need to learn when to rely on the fast engine versus the precision flagship when complex visual fidelity is on the line.