Overview
- Tool name: MAI-Image-2.5 Pro
- Developer: Microsoft AI (MAI)
- Official announcement: https://microsoft.ai/news/introducing-mai-image-2-5-pro-and-mai-voice-2-flash/
- Access: Microsoft Foundry (Azure AI Foundry)
- Category: AI image models
- Status: public preview
MAI-Image-2.5 Pro is the highest-fidelity image model in Microsoft's in-house MAI family, aimed at jobs where quality, consistency and creative control matter more than throughput. Microsoft stresses these models were trained in-house on clean, traceable, enterprise-grade data, without distillation from third-party models.
Official pricing (Microsoft Foundry)
| Billing item | MAI-Image-2.5 Pro | MAI-Image-2.5 |
|---|---|---|
| Text input | $5 per 1M tokens | $5 per 1M tokens |
| Image input | $8 per 1M tokens | $8 per 1M tokens |
| Image output | $106 per 1M tokens | $47 per 1M tokens |
Pricing verified against Microsoft's official announcement (microsoft.ai/news) and the Azure AI Foundry blog (October 2026).
Pro costs 2.26x the standard tier on image output. The logic is explicit: spend the compute on the handful of assets that need maximum fidelity, and run everything else on standard.
Capabilities
- Stronger object consistency: the same product, character or brand element stays consistent across multiple images
- Closer alignment with creative intent: enhanced visual reasoning and world knowledge, so camera direction, lighting and staging instructions land more accurately
- In-image text rendering: Microsoft names precise in-image text as a core capability
- Natural-language editing: describe the change and iterate
Pro or standard? (Microsoft's own guidance)
Microsoft gives four tests - hit any one and you want Pro:
1. You need maximum image fidelity and creative quality
2. Object consistency across multiple images is critical
3. Your workflow depends on visual reasoning, world knowledge and close adherence to creative direction
4. You are producing professional marketing, advertising, product design or enterprise creative assets
Scenarios Microsoft lists
- Campaign hero assets and multi-frame storytelling: agencies and marketing teams keep products, characters and brand elements consistent across launch campaigns, social variants, retail signage and digital advertising
- Product photography and e-commerce catalogs: accurate rendering of packaging, labels, materials and reflections, with identical product representation across angles, colorways and lifestyle settings
- Storyboarding and pre-visualization: film, game and creative production teams develop concepts fast while keeping characters, environments, props and scenes consistent
- Regulated-industry content: healthcare, financial services and manufacturing imagery that needs greater real-world accuracy and domain understanding, cutting correction effort before customer-facing use
Where it already runs (per Microsoft)
MAI models are running in production at scale across Microsoft products:
- Bing Image Creator is now 100% in-house by default, with MAI-Image-2.5 driving generation and editing end to end
- PowerPoint: image-to-image, with Microsoft reporting GPU costs up to 84% lower than GPT-Image-2
- OneDrive: default model for key image-editing scenarios, with Microsoft reporting 26% higher save rates, roughly 25% lower P95 latency and 2.5x greater efficiency under medium-utilization production workloads
- Also Dynamics 365 and Azure
Distribution: Microsoft Foundry, MAI Playground, PowerPoint, OneDrive Photos. Foundry API access is moving from private preview to public preview.
Read those percentages with care: they are Microsoft's own claims about its own products, and the announcement provides no independent test, sample size or full workload definition.
Limitations
- Public preview - API surface and pricing may change
- Closed source, no weights, no self-hosting; tied to the Azure / Foundry ecosystem
- Token-based billing is less predictable than per-image pricing (Ideogram 4.5, Grok Imagine 2.0); run a small batch before budgeting
- All production-efficiency figures are vendor-reported
Frequently asked questions
How much is MAI-Image-2.5 Pro?
Official rates: $5 per 1M text input tokens, $8 per 1M image input tokens and $106 per 1M image output tokens on Microsoft Foundry. Standard MAI-Image-2.5 is $47 per 1M image output tokens.
Pro or standard?
Pro when you need maximum fidelity, cross-image object consistency, visual reasoning and world knowledge, or you are producing professional marketing, advertising, product or enterprise creative assets. Standard is the better value otherwise.
Where can I use it?
Microsoft Foundry (Azure AI Foundry), MAI Playground, PowerPoint and OneDrive Photos; Bing Image Creator now defaults to MAI-Image-2.5. API access is in public preview.
Is it open source?
No - closed, served through Microsoft channels. Microsoft emphasizes the models were trained in-house on clean, traceable enterprise-grade data without distillation from third-party models.
MAI-Image-2.5 Pro vs GPT Image 2.5 vs Nano Banana Pro?
MAI-Image-2.5 Pro if you are in the Azure / Microsoft 365 ecosystem and need enterprise compliance and data residency; GPT Image 2.5 for the OpenAI ecosystem and strongest typography; Nano Banana Pro for the Google ecosystem and 4K.
*This page is not affiliated with Microsoft. Specs and pricing verified as of October 2026 (microsoft.ai announcement and the Azure AI Foundry blog); confirm current figures on the official site.*