MAI-Image-2.5
microsoft's own model, strong scores, awkward to buy
third on artificial analysis and genuinely excellent at text and stylised art — held back almost entirely by how hard it is to buy if you aren't already on azure.
released in june, mai-image-2.5 is microsoft's own image model rather than a rebadged openai one, and it is good: third on artificial analysis text-to-image, fourth on editing, sixth on lmarena. the jump over the previous generation was largest in text rendering and in cartoon, anime and fantasy styles.
its editing approach — microsoft calls it control with preservation — is aimed squarely at the failure mode where a targeted change quietly redraws the rest of the frame, and it handles that well.
the description above is ours, condensed from the ranking. pricing moves — check it on the vendor's own page before you rely on it.
- category
- ai image models
- pricing
- token-billed: $47 / 1M image-output tokens
- website
- microsoft.ai
third on artificial analysis and genuinely excellent at text and stylised art — held back almost entirely by how hard it is to buy if you aren't already on azure.
we researched this category against vendors' own pricing pages and licence files. that is where this line comes from — not from the vendor, and not from anything they paid for.
more ai image models
Leonardo.Ai
image generation platform with fine-tuned community models and a canvas editor
Google Imagen
google's text-to-image family, available through the gemini api and vertex ai