Microsoft Launches MAI-Image-2-Efficient: 41% Cheaper AI Image Generation

Microsoft Unveils Budget-Friendly AI Image Generator

Microsoft on Tuesday launched MAI-Image-2-Efficient, a streamlined version of its flagship text-to-image model designed to dramatically reduce the cost of AI-generated visuals for enterprise customers operating at scale.

The new model is priced at $19.50 per million image output tokens, compared with $33 for the full MAI-Image-2, representing a 41% cost reduction. It also runs 22% faster and achieves four times greater throughput efficiency per GPU, as measured on NVIDIA H100 hardware at 1024x1024 resolution.

Built for Volume, Not Perfection

MAI-Image-2-Efficient is aimed squarely at use cases where speed and scale outweigh pixel-level polish. Microsoft described the model as a "production powerhouse" for enterprises generating social media ads, placeholder mockups, product thumbnails, and batch marketing assets under tight budgets.

The flagship MAI-Image-2, which launched in March and climbed to third place on the Arena.ai leaderboard for text-to-image models, remains the company's choice for high-fidelity creative work such as cinematic scenes, hero portraits, and polished final graphics. Both models share the same photorealism and in-image text capabilities.

Immediate Availability

Developers can access MAI-Image-2-Efficient immediately through Microsoft Foundry and the MAI Playground with no waitlist. Text input is priced at $5 per million tokens, matching the rate of its premium sibling. Microsoft claims the model is also 40% faster than competing hyperscaler image models, including several from Google. A rollout to Copilot and Bing is planned, though no date has been specified.

Microsoft's Growing In-House AI Stack

The release extends a rapid cadence from Mustafa Suleyman's AI Superintelligence team, which shipped MAI-Image-1 in October 2025, MAI-Image-2 in March 2026, and earlier this month made MAI-Voice-1 and MAI-Transcribe-1 broadly available through Foundry. With each launch, Microsoft has moved further towards building its own model stack rather than relying on external partners such as OpenAI, a shift that now spans speech, voice, and image generation at competitive price points. The company has set a target of developing its own frontier AI models by 2027.