About MAI-Image-2.5
Overview
MAI-Image-2.5 is Microsoft AI's image generation and editing model, released on June 2, 2026. It ranks No. 2 on Arena's image-editing leaderboard and No. 3 for text-to-image, ahead of GPT-Image-1.5 and Nano Banana Pro 2K in the published benchmarks.
Microsoft ships the model in two forms: MAI-Image-2.5 for maximum fidelity and MAI-Image-2.5-Flash for fast, scalable production workloads. Developers reach both through Foundry, the MAI Playground, and OpenRouter, while Microsoft products apply them directly — PowerPoint for generating presentation visuals and OneDrive for precise photo edits.
Key Benefits
- Generates detailed, coherent images from prompts with stronger text rendering, product imagery, and prompt adherence.
- Makes edits that fit the scene by reasoning about structure, lighting, scale, and spatial relationships.
- Applies localized changes — swapping an object, updating text, or removing motion blur — without altering the rest of the frame.
- Keeps facial identity recognizable across edits that change pose, expression, or viewpoint.
- Scores a +75 point overall gain over MAI-Image-2, with the largest jumps in text rendering (+107) and cartoon, anime and fantasy (+90).
- Filters both prompts and outputs to detect and block harmful or policy-violating content.
How It Works
Users send a text prompt to either model through Foundry, the MAI Playground, or OpenRouter, then review the returned image. For edits, they name the specific element to change — an object, a line of text, or motion blur — and the model reworks that area while leaving the surrounding image intact. Teams pick the fidelity model when output quality matters and Flash when speed and cost drive the workload.
Use Cases
- Presentation designers generate slide-ready visuals and whole decks from prompts inside PowerPoint.
- Photo editors remove distractions and clean up backgrounds in OneDrive while preserving the original scene.
- Product marketers produce product imagery and in-image text that holds up in campaigns.
- Application developers call the model through Foundry or OpenRouter APIs to add generation and editing to their own tools.
- High-volume engineering teams run bulk editing jobs on MAI-Image-2.5-Flash to hold down output token spend.
Why Choose This Product
Microsoft pairs its Arena placement with published per-token pricing, so teams can weigh fidelity against cost before committing: MAI-Image-2.5 charges $5 per 1M text input tokens and $47 per 1M image output tokens, while Flash charges $1.75 and $19.50 on the same measures. The trade-offs are disclosed as well — the model can reflect training-data biases and may produce plausible but inaccurate visual details, so Microsoft recommends reviewing generated images before use in identity, legal, medical, financial, or news workflows.
MAI-Image-2.5 Pros & Cons
- Ranks No. 2 on Arena's image-editing leaderboard
- Ranks No. 3 for text-to-image on Arena
- +75 point overall gain over MAI-Image-2, with +107 in text rendering
- Fidelity and low-cost Flash variants share one API surface
- Live in PowerPoint and rolling out in OneDrive
- Can reflect biases present in training data
- May produce plausible but inaccurate or misleading visual details
- Microsoft advises review before use in identity, legal, medical, financial, or news workflows
Key Features
Text-to-Image Generation
Produces more detailed, coherent images from prompts with stronger text rendering, product imagery, and prompt adherence.
Visual Reasoning
Reads scene structure, lighting, scale, and spatial relationships so added objects land with the right perspective and shadows.
Fine-Grained Edits
Replaces an object, updates text, or removes motion blur in a localized area without changing the rest of the image.
Face Consistency
Preserves facial identity across edits and keeps likeness recognizable through changes in pose, expression, or viewpoint.
Fidelity and Flash Variants
MAI-Image-2.5 targets maximum fidelity while MAI-Image-2.5-Flash handles fast, scalable production workloads.
Safety Guardrails
Layered filtering screens both prompts and outputs to detect and block harmful or policy-violating content.
PowerPoint Image Generation
Generates presentation-ready visuals and slide content from prompts inside PowerPoint.
OneDrive Photo Editing
Rolls out precise photo edits in OneDrive, including removing distractions, cleaning backgrounds, and enhancing images.
Developer Access via Foundry
Ships to developers in Foundry and the MAI Playground, with OpenRouter distributing the model to its developer community.
MAI-Image-2.5 Pricing
- Faster, lower-cost generation and editing
- Text input tokens at $1.75 per 1M
- Image input tokens at $1.75 per 1M
- Image output tokens at $19.50 per 1M
- Suited to scalable production workloads
- Maximum fidelity output
- Text input tokens at $5 per 1M
- Image input tokens at $8 per 1M
- Image output tokens at $47 per 1M
- Fine-grained, controllable editing
| Feature | MAI-Image-2.5-Flash $1.75 usage-based | MAI-Image-2.5 $5 usage-based |
|---|---|---|
| Faster, lower-cost generation and editing | ||
| Text input tokens at $1.75 per 1M | ||
| Image input tokens at $1.75 per 1M | ||
| Image output tokens at $19.50 per 1M | ||
| Suited to scalable production workloads | ||
| Maximum fidelity output | ||
| Text input tokens at $5 per 1M | ||
| Image input tokens at $8 per 1M | ||
| Image output tokens at $47 per 1M | ||
| Fine-grained, controllable editing |
Pricing extracted from the product website and may change. Check the source for current details.
Frequently asked questions about MAI-Image-2.5
What is MAI-Image-2.5?
MAI-Image-2.5 is Microsoft AI's image model for high-quality generation and precise, controllable editing. It ranks No. 2 on Arena's image-editing leaderboard and No. 3 for text-to-image, and Microsoft positions it for production-ready image workflows in developer tools and Microsoft products.
What is the difference between MAI-Image-2.5 and MAI-Image-2.5-Flash?
MAI-Image-2.5 targets maximum fidelity, while MAI-Image-2.5-Flash targets fast, scalable production workloads at a lower cost. Flash charges $1.75 per 1M text and image input tokens and $19.50 per 1M image output tokens, against $5, $8, and $47 for the fidelity model.
How much does MAI-Image-2.5 cost?
In Foundry, MAI-Image-2.5 costs $5 per 1M text input tokens, $8 per 1M image input tokens, and $47 per 1M image output tokens. MAI-Image-2.5-Flash costs $1.75, $1.75, and $19.50 per 1M tokens on the same measures.
Where can developers access MAI-Image-2.5?
Both models are available to developers in Foundry, and users can try them in the MAI Playground. OpenRouter also makes MAI-Image-2.5 available to its developer community through the API those developers already use.
How does MAI-Image-2.5 perform on Arena benchmarks?
MAI-Image-2.5 ranks No. 2 on Arena's image-editing leaderboard, ahead of Nano Banana 2.1, and No. 3 for text-to-image, surpassing GPT-Image-1.5 and Nano Banana Pro 2K. It delivers a +75 point overall improvement over MAI-Image-2, with the largest gains in text rendering (+107) and cartoon, anime and fantasy (+90).
Does MAI-Image-2.5 keep faces consistent during edits?
Yes. The model preserves facial identity across edits and maintains a recognizable likeness even when pose, expression, or viewpoint changes.