Skip to main content

Microsoft AIAI ForeverImage model matchup

MAI-Image-2.5vs.Kandinsky 3.1

Which one is better?

Kandinsky 3.1 has no test of our own, no arena rating and no API price yet.

GraduallyBench Image v2Our own benchmark0 to 100 · higher is better
72

MAI-Image-2.5

Rank 29 of 108

n/a

Kandinsky 3.1

not tested yet

Arena ratingfrom votes by users · higher is better
1,253

MAI-Image-2.5

Rank 9 of 21

n/a

Kandinsky 3.1

not in the arena

Price per imagedollars through the API · lower is better
$0.048

MAI-Image-2.5

n/a

Kandinsky 3.1

Ranks among the 21 image models with an arena rating and the 108 with a GraduallyBench Image v2 run; prices compared with 28 models with an API price.

Test images from GraduallyBench Image v2

Our own benchmarkThe images come from GraduallyBench Image v2, our own benchmark: every model gets the same four tasks, and two AI judges compare the images in pairs and blind.

Same task, the first image of each.

MAI-Image-2.5: Typography and layout, first image from the fixed test run
MAI-Image-2.578 out of 100

No image from the four fixed tasks for Kandinsky 3.1 yet.

Kandinsky 3.1not tested yet

Result for “Typography and layout”:only MAI-Image-2.5 tested

Prompt given to both modelsCreate a square editorial poster for an imaginary night train called MONDFALTER. Show the exact German headline MONDFALTER and the exact subline BERLIN NACH LISSABON. Use a restrained midnight-blue and warm-cream palette, one stylized moth, strong Swiss-grid typography, generous negative space, and no additional words or logos.

What the rating looks at: Legible text, hierarchy, and composition.

“Ahead”: the 95 % ranges of both values do not overlap. “Slightly ahead”: a higher value, but the ranges overlap.

Missing image run for Kandinsky 3.1

FusionBrain timed out during our check. The public 3.1 repository demonstrates a standalone Flash pipeline, while the vendor distinguishes 3.1 from 3.1 Flash. This does not unambiguously establish execution of the complete regular 3.1 offering; Flash results are not substituted. Last checked November 2, 2026. Resolve verified access or the complete vendor pipeline for the regular 3.1 offering.

Documentation

Subscription and API

MAI-Image-2.5

Subscription
No Microsoft AI subscription access recorded. Through other providers from €16 a month (Magnific Premium, with limited credits). 6 plans in Germany.
Via API
$0.048 per image, 100 images $4.81.
Weights
Closed weights.

Kandinsky 3.1

Subscription
n/a No plan in Germany names exactly this model.
Via API
No API price known (access: local).
Weights
Open weights.

Subscriptions only count plans that name exactly this model. Subscription prices include VAT, API prices are in dollars.

Arena rating (LMArena) from Nov 19, 2026 · GraduallyBench Image v2 from Nov 7, 2026 · API prices from Nov 12, 2026 · Subscription database from Sep 29, 2026 (54 days old)

All data side by side

The better value is bold. For ratings only when the gap is clear.

KennzahlMAI-Image-2.5Kandinsky 3.1
Ratings
Arena ratingfrom votes by users, higher is better1,253n/a
Arena rankRank 9 of 21n/a
Arena votesmore votes, a safer rating58,997n/a
GraduallyBench Image v2our own benchmark, 0 to 10072n/a
GraduallyBench Image v2 by task
Typography and layoutLegible text, hierarchy, and composition78n/a
Product photographyMaterials, reflections, and fine details67n/a
Character and detailAnatomy, clothing, and spatial consistency79n/a
InfographicNumbers, labels, and visual organization63n/a
API price
Per imagelower is better$0.048n/a
100 images$4.81n/a
Profile
ReleasedJune 2026n/a
Weightsopen means: runs locallyproprietaryopen
LifecyclePreview in Microsoft Foundryno notice
AccessAPI, Web applocal
Parameters20 billionnot published
Largest output1,048,576 pixelsnot consistently published
Plans in Germanynaming exactly this model6n/a

API prices for one image at 1,024 × 1,024 pixels according to Artificial Analysis.

Sources: MAI-Image-2.5 model source, status source (verified Aug 15, 2026) · Kandinsky 3.1 model source (verified Nov 2, 2026)

Arena ratings by category

From blind votes on Artificial Analysis and LMArena; the sources are named under the card. “Ahead” means the lead is clear.

No shared arena score in the stored snapshot

The stored benchmark database contains no measurement for MAI-Image-2.5 and Kandinsky 3.1 from the same arena and category. Older model versions are not used as substitutes.

More comparisons

Compare another matchup