Image modelPreview

MAI-Image-2.5 Flash

Microsoft AI

Released: June 2026
Data date: September 22, 2026

Parameters: 20B
Access: API, Web
Benchmark configuration: Flash
Price: $0.02
Per image

MAI-Image-2.5 Flash is the faster preview sibling of MAI-Image-2.5. It likewise supports text-to-image and image-to-image editing, remains within 1,048,576 total pixels, and differs through its serving variant rather than a different image workflow.

Microsoft Foundry describes Flash as a faster, lower-cost preview sibling with the same generation and editing capabilities. The documentation does not call it a successor. MAI-Image-2.5 and Flash therefore remain separate sibling variants.

Position: Rank 3 of 30
Index score: 73 / 100

1.Krea 2 Large75.6 / 100

2.GPT Image 274.2 / 100

3.MAI-Image-2.5 Flash73 / 100

4.Krea 2 Medium Turbo65.2 / 100

5.Luma Uni-1.1 Max63.8 / 100

6.MAI-Image-2.561.8 / 100

7.GPT Image 2.5 Sunburst60.9 / 100

8.Reve 2.160.1 / 100

9.Seedream 5.0 Pro57.2 / 100

10.HiDream-O1-Image-1.556.3 / 100

11.GPT Image 2.5 Flare56.3 / 100

12.HunyuanImage 3.0 Instruct55.8 / 100

13.FLUX.2 Max55.8 / 100

14.Nano Banana Pro53.8 / 100

15.Nano Banana 2 Lite53.2 / 100

16.Recraft V4.1 Utility51.4 / 100

17.Nano Banana 251.1 / 100

18.FLUX.2 Flex49.4 / 100

19.GPT Image 1.548.3 / 100

20.Qwen Image 2.0 Pro47.7 / 100

21.Firefly Image Model 546 / 100

21.FLUX.2 Pro46 / 100

23.Ideogram 4.045.4 / 100

24.Wan 2.6 Text to Image41.4 / 100

25.Seedream 4.037.7 / 100

26.Grok Imagine Image Quality32.5 / 100

27.Midjourney V8.128.5 / 100

28.FLUX.2 Klein 9B24.4 / 100

29.HiDream-O1-Image19.3 / 100

30.Stable Diffusion 3.5 Large8.3 / 100

Methodology and source

This position gives equal weight to 4 fixed Gradually image tests. One archived first output contributes for each model and task.

Source: Gradually-Bildtest

Benchmark sources at a glance

Each evaluator measures different capabilities under its own conditions. Test series, composite indices and their components stay separate. The detailed results below retain their scales, configurations and source dates.

No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.

Coverage is not a quality ranking. Overlapping tests and multiple settings do not count as independent experiments. Vendor claims, historical and disputed results are excluded from this overview and remain labelled in the detailed evidence.

Published benchmarks

Text to Image Arena

Elo ratings summarize pairwise comparisons within this leaderboard. They are not percentages. The scale does not express absolute quality or linear gaps between models.

1.GPT Image 2.5 Sunburst1,197 Elo

2.GPT Image 2.5 Flare1,190 Elo

3.GPT Image 21,171 Elo

4.MAI-Image-2.61,147 Elo

5.Nano Banana 21,122 Elo

6.Muse Image1,112 Elo

7.MAI-Image-2.51,103 Elo

8.GPT Image 1.51,102 Elo

9.MAI-Image-2.6 Flash1,100 Elo

9.Nano Banana Pro1,100 Elo

11.Nano Banana 2 Lite1,092 Elo

12.Seedream 5.0 Pro1,078 Elo

13.Grok Imagine Image Quality1,043 Elo

14.MAI-Image-2.5 Flash1,033 Elo

15.Qwen Image 2.0 Pro1,030 Elo

16.FLUX.2 Flex1,026 Elo

16.Seedream 4.01,026 Elo

18.Krea 2 Large1,023 Elo

19.FLUX.2 Max1,022 Elo

19.HiDream-O1-Image-1.51,022 Elo

21.Recraft V4.1 Utility1,020 Elo

22.Krea 2 Medium Turbo1,016 Elo

23.Luma Uni-1.1 Max1,013 Elo

23.Wan 2.6 Text to Image1,013 Elo

25.Ideogram 4.01,004 Elo

26.FLUX.2 Pro1,001 Elo

27.HiDream-O1-Image979 Elo

28.HunyuanImage 3.0 Instruct964 Elo

29.FLUX.2 Klein 9B940 Elo

30.Stable Diffusion 3.5 Large839 Elo

higher is betterAs of September 22, 2026Scale: 0 Elo - 1,197 Elo

Source: Artificial Analysis

gradually.ai

Methodology and context

Metric: elo-rating
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 30
Sample: 5,742

Source: Artificial Analysis

Image Editing Arena

Elo ratings summarize pairwise comparisons within this leaderboard. They are not percentages. The scale does not express absolute quality or linear gaps between models.

1.GPT Image 2.5 Sunburst1,180 Elo

2.GPT Image 2.5 Flare1,161 Elo

3.MAI-Image-2.61,133 Elo

4.MAI-Image-2.6 Flash1,124 Elo

5.GPT Image 21,122 Elo

6.Muse Image1,116 Elo

7.MAI-Image-2.51,113 Elo

8.Seedream 5.0 Pro1,107 Elo

9.Nano Banana 21,106 Elo

10.GPT Image 1.51,103 Elo

11.Nano Banana Pro1,096 Elo

12.MAI-Image-2.5 Flash1,087 Elo

13.Grok Imagine Image Quality1,081 Elo

14.Luma Uni-1.1 Max1,068 Elo

15.Nano Banana 2 Lite1,042 Elo

16.HunyuanImage 3.0 Instruct1,027 Elo

17.Qwen Image 2.0 Pro1,016 Elo

18.FLUX.2 Klein 9B1,014 Elo

18.Seedream 4.01,014 Elo

20.FLUX.2 Pro1,008 Elo

21.FLUX.2 Flex1,006 Elo

22.FLUX.2 Max997 Elo

23.HiDream-O1-Image954 Elo

higher is betterAs of September 22, 2026Scale: 0 Elo - 1,180 Elo

Source: Artificial Analysis

gradually.ai

Methodology and context

Metric: elo-rating
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 23
Sample: 12,248

Source: Artificial Analysis

Gradually sample test: Typography and layout

The 0-100 value normalizes the mean rank from model-blind comparison rounds. One archived first output contributes per model. This remains an n = 1 sample.

1.Nano Banana 290.8 rank points

2.Nano Banana Pro89.7 rank points

3.Firefly Image Model 587.3 rank points

4.Grok Imagine Image Quality83.9 rank points

5.MAI-Image-2.582.8 rank points

6.Recraft V4.1 Utility79.3 rank points

7.Nano Banana 2 Lite78.2 rank points

8.Krea 2 Large74.7 rank points

9.HunyuanImage 3.0 Instruct69 rank points

10.Krea 2 Medium Turbo67.8 rank points

11.FLUX.2 Flex66.7 rank points

11.FLUX.2 Max66.7 rank points

13.Ideogram 4.058.6 rank points

14.HiDream-O1-Image-1.555.2 rank points

15.MAI-Image-2.5 Flash51.7 rank points

16.Luma Uni-1.1 Max50.6 rank points

17.FLUX.2 Pro49.4 rank points

18.Seedream 5.0 Pro48.3 rank points

19.GPT Image 241.4 rank points

20.Qwen Image 2.0 Pro40.2 rank points

21.GPT Image 2.5 Flare33.3 rank points

22.GPT Image 1.529.9 rank points

23.Reve 2.128.7 rank points

24.GPT Image 2.5 Sunburst21.8 rank points

25.Seedream 4.019.6 rank points

26.Wan 2.6 Text to Image13.8 rank points

27.FLUX.2 Klein 9B10.3 rank points

28.Midjourney V8.14.6 rank points

29.HiDream-O1-Image3.4 rank points

30.Stable Diffusion 3.5 Large2.3 rank points

higher is betterAs of September 28, 2026Scale: 0 rank points - 100 rank points

Source: Gradually-Bildtest

gradually.ai

Methodology and context

Metric: rank-score-100
Scale: 0-100
Participants: 30
Sample: 1
Methodology and context: One archived, unfiltered first output per model. All 30 images are ranked in three model-blind orderings. The 0-100 value normalizes the mean rank. This remains an n = 1 sample, and an automated judge may have visual preferences of its own.

Source: Gradually-Bildtest

Gradually sample test: Product photography

The 0-100 value normalizes the mean rank from model-blind comparison rounds. One archived first output contributes per model. This remains an n = 1 sample.

1.GPT Image 2.5 Sunburst97.7 rank points

2.Krea 2 Medium Turbo88.5 rank points

3.GPT Image 283.9 rank points

4.Krea 2 Large81.6 rank points

4.MAI-Image-2.5 Flash81.6 rank points

6.MAI-Image-2.579.3 rank points

7.Luma Uni-1.1 Max73.6 rank points

8.Reve 2.171.3 rank points

9.Seedream 5.0 Pro70.1 rank points

10.Nano Banana 2 Lite65.5 rank points

11.Qwen Image 2.0 Pro57.5 rank points

12.FLUX.2 Max56.3 rank points

13.HunyuanImage 3.0 Instruct55.2 rank points

14.FLUX.2 Flex52.9 rank points

15.FLUX.2 Pro51.7 rank points

15.Nano Banana 251.7 rank points

15.Nano Banana Pro51.7 rank points

18.FLUX.2 Klein 9B43.7 rank points

18.GPT Image 1.543.7 rank points

18.GPT Image 2.5 Flare43.7 rank points

21.Midjourney V8.137.9 rank points

22.Firefly Image Model 533.3 rank points

22.HiDream-O1-Image-1.533.3 rank points

24.Recraft V4.1 Utility28.7 rank points

25.Seedream 4.027.6 rank points

26.Wan 2.6 Text to Image14.9 rank points

27.Ideogram 4.012.7 rank points

28.HiDream-O1-Image5.8 rank points

29.Grok Imagine Image Quality4.6 rank points

30.Stable Diffusion 3.5 Large0 rank points

higher is betterAs of September 28, 2026Scale: 0 rank points - 100 rank points

Source: Gradually-Bildtest

gradually.ai

Methodology and context

Metric: rank-score-100
Scale: 0-100
Participants: 30
Sample: 1
Methodology and context: One archived, unfiltered first output per model. All 30 images are ranked in three model-blind orderings. The 0-100 value normalizes the mean rank. This remains an n = 1 sample, and an automated judge may have visual preferences of its own.

Source: Gradually-Bildtest

Gradually sample test: Character and detail

The 0-100 value normalizes the mean rank from model-blind comparison rounds. One archived first output contributes per model. This remains an n = 1 sample.

1.GPT Image 293.1 rank points

2.Wan 2.6 Text to Image88.5 rank points

3.GPT Image 2.5 Flare87.3 rank points

4.MAI-Image-2.5 Flash82.8 rank points

5.GPT Image 1.577 rank points

6.Luma Uni-1.1 Max75.9 rank points

7.FLUX.2 Max71.3 rank points

8.HunyuanImage 3.0 Instruct69 rank points

9.Firefly Image Model 563.2 rank points

10.HiDream-O1-Image62.1 rank points

11.FLUX.2 Pro60.9 rank points

12.HiDream-O1-Image-1.559.8 rank points

12.Midjourney V8.159.8 rank points

12.Recraft V4.1 Utility59.8 rank points

15.Krea 2 Large56.3 rank points

16.FLUX.2 Flex54 rank points

17.Qwen Image 2.0 Pro48.3 rank points

18.Reve 2.142.5 rank points

19.Nano Banana Pro39.1 rank points

20.Krea 2 Medium Turbo29.9 rank points

20.Seedream 4.029.9 rank points

22.FLUX.2 Klein 9B28.7 rank points

23.Stable Diffusion 3.5 Large26.4 rank points

24.GPT Image 2.5 Sunburst25.3 rank points

25.Ideogram 4.023 rank points

26.MAI-Image-2.520.7 rank points

27.Grok Imagine Image Quality18.4 rank points

27.Seedream 5.0 Pro18.4 rank points

29.Nano Banana 217.2 rank points

30.Nano Banana 2 Lite11.5 rank points

higher is betterAs of September 28, 2026Scale: 0 rank points - 100 rank points

Source: Gradually-Bildtest

gradually.ai

Methodology and context

Metric: rank-score-100
Scale: 0-100
Participants: 30
Sample: 1
Methodology and context: One archived, unfiltered first output per model. All 30 images are ranked in three model-blind orderings. The 0-100 value normalizes the mean rank. This remains an n = 1 sample, and an automated judge may have visual preferences of its own.

Source: Gradually-Bildtest

Gradually sample test: Infographic

The 0-100 value normalizes the mean rank from model-blind comparison rounds. One archived first output contributes per model. This remains an n = 1 sample.

1.GPT Image 2.5 Sunburst98.9 rank points

2.Reve 2.197.7 rank points

3.Seedream 5.0 Pro92 rank points

4.Krea 2 Large89.7 rank points

5.Ideogram 4.087.3 rank points

6.GPT Image 278.2 rank points

7.HiDream-O1-Image-1.577 rank points

8.MAI-Image-2.5 Flash75.9 rank points

9.Krea 2 Medium Turbo74.7 rank points

10.Seedream 4.073.6 rank points

11.MAI-Image-2.564.4 rank points

12.GPT Image 2.5 Flare60.9 rank points

13.Nano Banana 2 Lite57.5 rank points

14.Luma Uni-1.1 Max55.2 rank points

15.Wan 2.6 Text to Image48.3 rank points

16.Nano Banana 244.8 rank points

16.Qwen Image 2.0 Pro44.8 rank points

18.GPT Image 1.542.5 rank points

19.Recraft V4.1 Utility37.9 rank points

20.Nano Banana Pro34.5 rank points

21.HunyuanImage 3.0 Instruct29.9 rank points

22.FLUX.2 Max28.7 rank points

23.FLUX.2 Flex24.1 rank points

24.Grok Imagine Image Quality23 rank points

25.FLUX.2 Pro21.8 rank points

26.FLUX.2 Klein 9B14.9 rank points

27.Midjourney V8.111.5 rank points

28.HiDream-O1-Image5.8 rank points

29.Stable Diffusion 3.5 Large4.6 rank points

30.Firefly Image Model 50 rank points

higher is betterAs of September 28, 2026Scale: 0 rank points - 100 rank points

Source: Gradually-Bildtest

gradually.ai

Methodology and context

Metric: rank-score-100
Scale: 0-100
Participants: 30
Sample: 1
Methodology and context: One archived, unfiltered first output per model. All 30 images are ranked in three model-blind orderings. The 0-100 value normalizes the mean rank. This remains an n = 1 sample, and an automated judge may have visual preferences of its own.

Source: Gradually-Bildtest

Recorded measurement details

Text to Image Arena (Show measurement, test conditions, and source)

Source value: 1033
Score: 1033
Metric: elo-rating
Unit: Elo
Direction: Higher is better
Source rank: 18
Sample: 5742
Confidence interval: 9
Source type: independent-evaluator
Evaluator: Artificial Analysis
Retrieved at: 2026-09-22
USD price per 1,000 tokens: 20

Source Artificial Analysis

Image Editing Arena (Show measurement, test conditions, and source)

Source value: 1087
Score: 1087
Metric: elo-rating
Unit: Elo
Direction: Higher is better
Source rank: 14
Sample: 12248
Confidence interval: 9
Source type: independent-evaluator
Evaluator: Artificial Analysis
Retrieved at: 2026-09-22
USD price per 1,000 tokens: 20

Source Artificial Analysis

Gradually sample test: Typography and layout (Show measurement, test conditions, and source)

Source value: 51.7
Score: 51.7
Metric: rank-score-100
Unit: rank score
Direction: Higher is better
Source rank: 15
Sample: 1
Source type: gradually-sample-test
Evaluator: gemini-3.1-pro-preview
Retrieved at: 2026-09-28
Methodology: One archived, unfiltered first output per model. All 30 images are ranked in three model-blind orderings. The 0-100 value normalizes the mean rank. This remains an n = 1 sample, and an automated judge may have visual preferences of its own.
Mean rank: 15
Individual ranks: 13, 14, 18
Criterion scores: headline 1.67, subline 1.67, text-control 1.67, visual-brief 1.67, layout 1

Source Gradually-Bildtest

Gradually sample test: Product photography (Show measurement, test conditions, and source)

Source value: 81.6
Score: 81.6
Metric: rank-score-100
Unit: rank score
Direction: Higher is better
Source rank: 4
Sample: 1
Source type: gradually-sample-test
Evaluator: gemini-3.1-pro-preview
Retrieved at: 2026-09-28
Methodology: One archived, unfiltered first output per model. All 30 images are ranked in three model-blind orderings. The 0-100 value normalizes the mean rank. This remains an n = 1 sample, and an automated judge may have visual preferences of its own.
Mean rank: 6.33
Individual ranks: 1, 11, 7
Criterion scores: subject 1.67, time 1.67, watch-details 1.67, photography 1.67, exclusions 1.67

Source Gradually-Bildtest

Gradually sample test: Character and detail (Show measurement, test conditions, and source)

Source value: 82.8
Score: 82.8
Metric: rank-score-100
Unit: rank score
Direction: Higher is better
Source rank: 4
Sample: 1
Source type: gradually-sample-test
Evaluator: gemini-3.1-pro-preview
Retrieved at: 2026-09-28
Methodology: One archived, unfiltered first output per model. All 30 images are ranked in three model-blind orderings. The 0-100 value normalizes the mean rank. This remains an n = 1 sample, and an automated judge may have visual preferences of its own.
Mean rank: 6
Individual ranks: 14, 3, 1
Criterion scores: scene 1, wardrobe 1, anatomy 1, props 1, environment 1

Source Gradually-Bildtest

Gradually sample test: Infographic (Show measurement, test conditions, and source)

Source value: 75.9
Score: 75.9
Metric: rank-score-100
Unit: rank score
Direction: Higher is better
Source rank: 8
Sample: 1
Source type: gradually-sample-test
Evaluator: gemini-3.1-pro-preview
Retrieved at: 2026-09-28
Methodology: One archived, unfiltered first output per model. All 30 images are ranked in three model-blind orderings. The 0-100 value normalizes the mean rank. This remains an n = 1 sample, and an automated judge may have visual preferences of its own.
Mean rank: 8
Individual ranks: 10, 6, 8
Criterion scores: title 1.67, sequence 1.67, labels 1.67, diagram 1.33, clarity 1.33

Source Gradually-Bildtest

Image samples from our tests

These images come from documented single requests with an archived prompt. They complement measurements but do not replace a controlled laboratory test.

Open Character and detail at original size

Character and detail

This sample tests Anatomy, clothing, and spatial consistency.

Prompt and settings

Create a square cinematic full-body portrait of a fictional bicycle courier waiting under a transparent umbrella at a rainy tram stop in Hamburg at blue hour. She wears a mustard raincoat, navy trousers, red sneakers, a silver helmet, and carries a teal messenger bag. Keep both hands visible, render the bicycle correctly, use realistic wet-street reflections, and include no readable brand names.

Generated through manual. Archived settings were Official Bing Image Creator web app; MAI-Image-2.5-Flash; one request; 1:1; first result; free guest generation; 2026-08-14; https://www.bing.com/images/create/ai-image-generator.

Open source

Open Infographic at original size

Infographic

This sample tests Numbers, labels, and visual organization.

Prompt and settings

Create a square German infographic titled SO FUNKTIONIERT PHOTOSYNTHESE. Show exactly four numbered steps in this order: 1 LICHT, 2 WASSER, 3 CO₂, 4 ZUCKER + SAUERSTOFF. Use a clean editorial science style with one plant cross-section, simple arrows, high contrast, legible labels, and no other text.

Open source

Open Product photography at original size

Product photography

This sample tests Materials, reflections, and fine details.

Prompt and settings

Create a square premium product photograph of a brushed titanium wristwatch standing upright on dark green marble. The watch has a cream dial, thin black hands set to 10:10, twelve distinct hour markers, a realistic crown, and a dark brown leather strap. Soft window light from the left, controlled reflections, shallow depth of field, no text, no logo, no extra objects.

Open source

Open Typography and layout at original size

Typography and layout

This sample tests Legible text, hierarchy, and composition.

Prompt and settings

Create 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.

Open source

Specifications and access

Specification	Value and source
Weights	ProprietarySource
Access	API, WebSource
Parameters	20BSource
Maximum output	1,048,576 pixelsSource
Benchmark configuration	FlashSource
Lifecycle	Preview in Microsoft FoundrySource

Sources and data date

Type	Evidence and data date
Additional source	Microsoft AI · Microsoft Foundry MAI image models documentation · Editorial description reviewed October 2, 2026
Additional source	Image pricing (retrieved September 22, 2026)
Additional source	Artificial Analysis (retrieved September 22, 2026)
Additional source	Gradually-Bildtest (retrieved September 28, 2026)
Additional source	Microsoft MAI-Image-2.5 model card · Editorial description reviewed October 2, 2026
Additional source	OpenRouter: Popularity index (derived), CC BY 4.0
Additional source	Vercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional source	Ollama: cumulative downloads for the popularity index
Additional source	Replicate: cumulative runs for the popularity index
Additional source	Google Trending Now: limited feed selection for Germany and the US
Additional source	Creative Commons Attribution 4.0

MAI-Image-2.5 Flash

Benchmark sources at a glance

Published benchmarks

Text to Image Arena

Image Editing Arena

Gradually sample test: Typography and layout

Gradually sample test: Product photography

Gradually sample test: Character and detail

Gradually sample test: Infographic

Recorded measurement details

Image samples from our tests

Character and detail

Infographic

Product photography

Typography and layout

Popularity

Specifications and access

Published prices

Compare this model

Sources and data date