Microsoft launched a new AI image model today, and the most interesting thing about it isn't the technology. It's the price.
MAI-Image-2-Efficient is a production-focused variant of Microsoft's flagship MAI-Image-2 model. It's available right now in Microsoft Foundry and the MAI Playground, with no waitlist. And compared to the flagship, it costs 41% less per image generated.
Here's the quick breakdown of what changed:
- Image output pricing: $19.50 per million image output tokens (down from $33)
- Speed: 22% faster than the flagship model
- Throughput: 4x greater efficiency per GPU
- Latency: Beats Google's competing Gemini image models by roughly 40% on standard benchmarks
That's not incremental. That's a meaningful shift in what it costs to run image generation at any kind of volume.
What This Model Is Actually Built For
Microsoft is being unusually clear about who this model is for and what it should be used for. The company explicitly says MAI-Image-2-Efficient is designed for: product shots, marketing creative, UI mockups, branded assets, and batch pipelines.
That is a list written for small and mid-size businesses.
If you run an e-commerce store and you need product photography at scale, this is relevant. If you're a solo marketer producing social content every week, this is relevant. If you have a retail brand that needs to generate consistent visual assets across platforms without paying a designer for every variation, this is relevant.
The flagship MAI-Image-2 still exists for high-complexity work: photorealistic portraits, complex stylization, illustration, long in-image text. But for the everyday volume use cases that make up most of what small businesses actually need from AI images, Microsoft says the efficient model handles them cleanly.
The Pricing Shift That Matters
AI image generation has been getting cheaper for two years. But 41% off the flagship in under a month after launch is fast, even by current standards.
MAI-Image-2 itself only launched on March 19. MAI-Image-2-Efficient arrived today, April 14, less than four weeks later. Microsoft's own AI model team shipped a production-optimized variant in under a month. That pace signals something: this is an area Microsoft is actively pushing, not coasting on.
The pricing model also mirrors what's worked elsewhere in the AI industry. OpenAI has GPT-4o Mini alongside GPT-4o. Anthropic has Haiku alongside Claude Sonnet. Google has Gemini Flash. The pattern is consistent: a high-end model for precision work, a faster and cheaper model for production volume. Microsoft is now applying that same logic to image generation.
For small businesses, this means the tier you want to pay attention to is rarely the flagship. The efficient, cheaper, faster model is usually the one that maps to your actual workload.
How to Access It
MAI-Image-2-Efficient is available through two channels right now:
- MAI Playground at playground.microsoft.ai, no waitlist
- Microsoft Foundry (formerly Azure AI Foundry), for businesses already using Azure infrastructure
Microsoft also confirmed the model is rolling out across Copilot and Bing, which means you may encounter it soon without having to do anything at all.
If you want to test it deliberately, the MAI Playground is the fastest path. No account setup beyond a Microsoft login.
The Bigger Picture
What's happening here isn't just about one model or one price cut. It's about the overall direction of AI image generation costs.
Every time a major provider releases a more efficient model, the floor for what's acceptable pricing shifts. The $33 per million tokens that MAI-Image-2 launched at in April looks expensive when a model of similar quality is now available at $19.50. That pressure flows down to every provider in the market.
For small business owners who've been watching AI image tools but waiting for the cost to make sense, the window is closing on the "too expensive" argument. Across the board, from Microsoft to Google to OpenAI, the per-image economics are moving in your direction.
The question worth asking this week: what's on your current marketing budget for photography and visual asset creation, and what would it cost to run a 30-day test with one of these tools instead? The math may surprise you.
Sources: VentureBeat, April 14, 2026; The Verge, April 14, 2026; Microsoft MAI Playground