GPT Image 2.6 vs MAI-Image-2.6: two different providers
The shared version number does not mean the same model. Compare access, editing workflow and deliverables before choosing.
Editorial comparison, not a broad model benchmark. Unless explicitly labeled as GPT model test results, images are existing prompt-library examples with unverified original settings. The 2.5 vs 2.6 portrait section includes two new matched-prompt GPT Image outputs and their request records.
Official data & independent evaluations
Official specifications
Scroll horizontally to see every column ↔
| Measure | GPT Image 2.6 API / this studio | MAI-Image-2.6 · Foundry preview | Source |
|---|---|---|---|
| Model developer | OpenAI | Microsoft AI | Microsoft Learn · MAI image models GPT Image 2.6 browser generator |
| Reference inputs | Up to 4 in this studio | Up to 5 JPEG/PNG references in the edits API | Microsoft Learn · MAI image models GPT Image 2.6 browser generator |
| Published output constraints | PNG, GIF and SVG available here | PNG; maximum 2,359,296 pixels; each side at least 768 px | Microsoft Learn · MAI image models GPT Image 2.6 browser generator |
| Grounded generation | Not exposed by this studio | web_grounding uses Bing Search | Microsoft Learn · MAI image models GPT Image 2.6 browser generator |
| Deployment status | Browser access here; API access is separate | Foundry preview, no preview SLA | Microsoft Learn · MAI image models GPT Image 2.6 browser generator |
Arena · text-to-image
Scroll horizontally to see every column ↔
| Model / tested setting | Score ± interval | Votes / samples | Status |
|---|---|---|---|
| GPT-Image-2.6 Sunburst (Medium) | 1409 ±13 | 2,967 | Preliminary |
| MAI-Image-2.6 | 1331 ±5 | 38,071 | Listed |
Arena · single-image editing
Scroll horizontally to see every column ↔
| Model / tested setting | Score ± interval | Votes / samples |
|---|---|---|
| GPT-Image-2.6 Sunburst (Medium) | 1476 ±10 | 3,833 |
| MAI-Image-2.6 | 1426 ±4 | 89,003 |
Displayed snapshot: November 2, 2026; inspected November 6. Its cutoff predates the public 2.6 release, so the update alignment is unclear. Treat these as displayed historical values, not a verified November 6 ranking. Quality labels and vote counts differ.
Arena measures human preferences in head-to-head comparisons. Its scores are not percentages of image quality. Text-to-image and editing boards measure different tasks. Arena · rankings and confidence intervals
Artificial Analysis · AA-Image-T2I v2.1
The inspected AA board has no GPT Image 2.6 entry. Its Elo scale differs from Arena; do not subtract or combine scores across them. AA uses blind human comparisons, a 1:1 aspect ratio and seed 42 where supported.
Artificial Analysis · AA-Image-T2I v2.1 · Artificial Analysis · image methodology
What the evidence supports
Arena’s displayed scores favor 2.6 Sunburst Medium over MAI-Image-2.6 in these tasks. AA lists MAI at 1168 ±12 (8,500 samples), but has no GPT Image 2.6 entry: it cannot independently confirm that pairwise conclusion.
The practical difference is the interface contract: Foundry returns PNG and caps area, while this studio exposes motion and vectors. A 2048×2048 MAI request exceeds its published area limit; 1536×1536 reaches that limit. This is a delivery constraint, not an artistic quality score.
This studio at a glance
| Compare | GPT Image 2.6 | MAI-Image-2.6 |
|---|---|---|
| Provider | OpenAI | Microsoft AI |
| Available in this studio | Yes: Flare and Sunburst | No |
| Where to check access | Use this browser studio or OpenAI documentation | Microsoft model page and deployment documentation |
| What to compare | Required file and repeat-edit workflow | Preservation of the original subject in your own tests |
The lab name matters more than “2.6”
GPT Image and MAI-Image belong to separate model families. A Microsoft benchmark result cannot be assigned to OpenAI’s model. Confirm the full model name, provider and test date whenever you read a comparison. This studio provides only the OpenAI variants.
For edits, write down what must stay
Use a photo you are allowed to upload. Ask for one local change and list the details to preserve: camera, object shape, label and surrounding layout. Judge the unchanged regions as carefully as the requested edit. A shared task makes a useful comparison; unrelated showcase images do not.
Choose access and export together
Microsoft publishes its own access and deployment information. Verify the exact offering before assuming an API feature is available in a consumer interface. Here, PNG, GIF and SVG share the same 20-credit run cost; this is this website’s pricing, not Microsoft or OpenAI API pricing.
Questions before you choose
Is MAI-Image-2.6 another name for GPT Image 2.6?
No. MAI-Image is Microsoft’s family; GPT Image is OpenAI’s.
Does a leaderboard settle the choice?
Only for its tested task, versions and date. Run your own editing brief before adopting a model for production.
Sources & scope
Try your own brief
One free generation on sign-in. Successful runs cost 20 credits; failed runs restore their credits.
Go to the homepage and start creating →