Microsoft AIImage modelPreview

MAI-Image-2.5

MAI-Image-2.5 is rank 9 of 21 in the arena rating and rank 26 of 108 in our GraduallyBench test.

Arena ratingRank 9 of 21

1,253User voteshigher is better

GraduallyBenchRank 26 of 108

69our own rating0 to 100

PriceRank 12 of 28

$0.048per imagelower is better

PopularityRank 221 of 875

43out of 100same as last month

Summary

At $0.048 per image it is fairly cheap; a typical image model costs $0.06.

MAI-Image-2.5 combines text-to-image and image-to-image editing in a diffusion architecture with 20 billion parameters. Its model card lists a 32K context, a 1,048,576-pixel ceiling (roughly 1,024 × 1,024), and text and image inputs.

Access is available as a preview through Microsoft Foundry. Microsoft notes that preview models can have limited capabilities and no production guarantees. The model card also reports a quality jump over MAI-Image-2 based on a provider comparison.

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 openrouter with model ID microsoft/mai-image-2.5. Archived settings were aspect ratio 1:1, provider default 1K resolution, n=1, seed=provider random, actual cost $0.048508.

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.

Generated through openrouter with model ID microsoft/mai-image-2.5. Archived settings were aspect ratio 1:1, provider default 1K resolution, n=1, seed=provider random, actual cost $0.048538.

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.

Generated through openrouter with model ID microsoft/mai-image-2.5. Archived settings were aspect ratio 1:1, provider default 1K resolution, n=1, seed=provider random, actual cost $0.048513.

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.

Generated through openrouter with model ID microsoft/mai-image-2.5. Archived settings were aspect ratio 1:1, provider default 1K resolution, n=1, seed=provider random, actual cost $0.048473.

Open source

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 50 published benchmark results

Published benchmarks

Text to image

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.

GPT Image 2.5 Sunburst1,227.25 Elo

GPT Image 2.5 Flare1,208.16 Elo

GPT Image 21,187.61 Elo

MAI-Image-2.61,168.32 Elo

Grok Imagine Image 2.01,165.79 Elo

Nano Banana 21,141.6 Elo

Muse Image1,127.86 Elo

GPT Image 1.51,123.86 Elo

MAI-Image-2.6 Flash1,121.56 Elo

MAI-Image-2.51,120.06 Elo

Nano Banana Pro1,118.67 Elo

Nano Banana 2 Lite1,109.07 Elo

Seedream 5.0 Pro1,101.17 Elo

Grok Imagine Image Quality1,049.08 Elo

Qwen-Image-2.11,045.48 Elo

Krea 2 Large1,035.45 Elo

MAI-Image-2.5 Flash1,032.28 Elo

Qwen Image 2.0 Pro1,031.23 Elo

FLUX.2 Flex1,027.68 Elo

Recraft V4.1 Utility1,027.66 Elo

FLUX.2 Max1,020.57 Elo

Krea 2 Medium Turbo1,020.35 Elo

Wan 2.6 Text to Image1,017.69 Elo

HiDream-O1-Image-1.51,017.68 Elo

Ideogram 4.51,015.99 Elo

FLUX.2 Pro1,003.17 Elo

Ideogram 4.01,002.93 Elo

HunyuanImage 3.0 Instruct988.05 Elo

HiDream-O1-Image978 Elo

FLUX.2 Klein 9B932.08 Elo

higher is betterAs of November 6, 2026Scale: 0 Elo - 1,227.25 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
Methodology and context: AA appearances: 14060. This count does not establish an independent sample. Lower rank: 8. Upper rank: 15. Source model ID: 3f8a2d6b-9c4e-4f1a-b7d2-8e5a0c3d71f9. Evidence capture ID: aa-image-text-to-image-v2.1-2026-11-06-e430d57b5d0b. Source payload hash: e430d57b5d0bf260d6feb60bb7d8fa32da8d7961ec2bd9a9cae58fb015613cf0

Source: Artificial Analysis

LMArena Text to Image: 3D modeling

GPT Image 2.5 Sunburst1,412.39 points

GPT Image 2.5 Flare1,381.8 points

GPT Image 21,365.33 points

MAI-Image-2.61,323.82 points

Grok Imagine Image 2.01,291.41 points

Reve 2.11,281.14 points

Nano Banana 21,255.68 points

MAI-Image-2.51,251.47 points

Seedream 5.0 Pro1,246.55 points

Muse Image1,238.45 points

Nano Banana Pro1,232.1 points

Nano Banana 2 Lite1,225.65 points

GPT Image 1.51,217.68 points

Qwen-Image-2.11,213.88 points

FLUX.2 Flex1,173.15 points

FLUX.2 Max1,160.58 points

FLUX.2 Pro1,145.47 points

HiDream-O1-Image1,120.68 points

Krea 2 Large1,115.65 points

FLUX.2 Klein 9B1,078.61 points

Stable Diffusion 3.5 Large963.03 points

higher is betterAs of November 6, 2026Scale: 0 points - 1,412.39 points

Source: LMArena

gradually.ai

Methodology and context

Metric: arena-score
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 21
Sample: 5,681
Methodology and context: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source: LMArena

LMArena Text to Image: Art

GPT Image 2.5 Sunburst1,410.08 points

GPT Image 2.5 Flare1,387.35 points

GPT Image 21,370.68 points

MAI-Image-2.61,337.7 points

Grok Imagine Image 2.01,328.25 points

Reve 2.11,290.77 points

MAI-Image-2.51,276.76 points

Seedream 5.0 Pro1,258.85 points

Muse Image1,258.79 points

Nano Banana 2 Lite1,255.8 points

Nano Banana 21,239.62 points

Qwen-Image-2.11,234.74 points

GPT Image 1.51,224.81 points

Nano Banana Pro1,224.23 points

FLUX.2 Max1,168.74 points

FLUX.2 Pro1,159.34 points

FLUX.2 Flex1,155.3 points

Krea 2 Large1,122.19 points

HiDream-O1-Image1,116.78 points

FLUX.2 Klein 9B1,078.63 points

Stable Diffusion 3.5 Large997.29 points

higher is betterAs of November 6, 2026Scale: 0 points - 1,410.08 points

Source: LMArena

gradually.ai

Methodology and context

Metric: arena-score
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 21
Sample: 9,115
Methodology and context: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source: LMArena

LMArena Text to Image: Cartoon

GPT Image 2.5 Sunburst1,456.88 points

GPT Image 2.5 Flare1,427.12 points

GPT Image 21,402.93 points

Grok Imagine Image 2.01,352.36 points

MAI-Image-2.61,348.71 points

Reve 2.11,309.78 points

Muse Image1,274.79 points

MAI-Image-2.51,268.65 points

Seedream 5.0 Pro1,263.52 points

Nano Banana 21,260.23 points

Nano Banana 2 Lite1,256.09 points

GPT Image 1.51,243.95 points

Nano Banana Pro1,243.91 points

Qwen-Image-2.11,234.48 points

FLUX.2 Max1,167.57 points

FLUX.2 Flex1,163.29 points

FLUX.2 Pro1,158.62 points

Krea 2 Large1,118.81 points

HiDream-O1-Image1,118.12 points

FLUX.2 Klein 9B1,063.58 points

Stable Diffusion 3.5 Large945.23 points

higher is betterAs of November 6, 2026Scale: 0 points - 1,456.88 points

Source: LMArena

gradually.ai

Methodology and context

Metric: arena-score
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 21
Sample: 24,138
Methodology and context: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source: LMArena

LMArena Text to Image: Commercial design

GPT Image 2.5 Sunburst1,415.51 points

GPT Image 21,385.8 points

GPT Image 2.5 Flare1,384.8 points

MAI-Image-2.61,332.9 points

Grok Imagine Image 2.01,324.96 points

Reve 2.11,315.65 points

Nano Banana 21,267.93 points

MAI-Image-2.51,259.36 points

Muse Image1,258.15 points

Nano Banana 2 Lite1,248.4 points

Nano Banana Pro1,245.44 points

GPT Image 1.51,237.88 points

Qwen-Image-2.11,237.36 points

Seedream 5.0 Pro1,236.31 points

FLUX.2 Flex1,172.44 points

FLUX.2 Max1,163.75 points

FLUX.2 Pro1,154.5 points

HiDream-O1-Image1,129.54 points

Krea 2 Large1,110.71 points

FLUX.2 Klein 9B1,070.2 points

Stable Diffusion 3.5 Large936.77 points

higher is betterAs of November 6, 2026Scale: 0 points - 1,415.51 points

Source: LMArena

gradually.ai

Methodology and context

Metric: arena-score
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 21
Sample: 22,289
Methodology and context: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source: LMArena

LMArena Text to Image: Overall

GPT Image 2.5 Sunburst1,427.13 points

GPT Image 2.5 Flare1,399.44 points

GPT Image 21,382.1 points

MAI-Image-2.61,330.58 points

Grok Imagine Image 2.01,317.31 points

Reve 2.11,300.4 points

Muse Image1,268.12 points

Nano Banana 21,257.49 points

MAI-Image-2.51,253 points

Seedream 5.0 Pro1,250.8 points

Nano Banana 2 Lite1,249.44 points

Nano Banana Pro1,246.31 points

GPT Image 1.51,236.39 points

Qwen-Image-2.11,226.55 points

FLUX.2 Max1,161.46 points

FLUX.2 Flex1,156.26 points

FLUX.2 Pro1,152.68 points

Krea 2 Large1,115.37 points

HiDream-O1-Image1,114.71 points

FLUX.2 Klein 9B1,069.84 points

Stable Diffusion 3.5 Large938.35 points

higher is betterAs of November 6, 2026Scale: 0 points - 1,427.13 points

Source: LMArena

gradually.ai

Methodology and context

Metric: arena-score
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 21
Sample: 61,222
Methodology and context: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source: LMArena

LMArena Text to Image: Photorealism

GPT Image 2.5 Sunburst1,434.4 points

GPT Image 2.5 Flare1,415.95 points

GPT Image 21,379.13 points

MAI-Image-2.61,330.88 points

Grok Imagine Image 2.01,312.19 points

Reve 2.11,302.1 points

Muse Image1,278.09 points

Seedream 5.0 Pro1,274.42 points

Nano Banana 21,261.31 points

Nano Banana Pro1,260.39 points

Nano Banana 2 Lite1,258.64 points

MAI-Image-2.51,250.67 points

GPT Image 1.51,246.79 points

Qwen-Image-2.11,229.66 points

FLUX.2 Max1,160.87 points

FLUX.2 Pro1,151.76 points

FLUX.2 Flex1,145.47 points

Krea 2 Large1,128.88 points

HiDream-O1-Image1,109.06 points

FLUX.2 Klein 9B1,091.05 points

Stable Diffusion 3.5 Large928.41 points

higher is betterAs of November 6, 2026Scale: 0 points - 1,434.4 points

Source: LMArena

gradually.ai

Methodology and context

Metric: arena-score
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 21
Sample: 23,657
Methodology and context: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source: LMArena

LMArena Text to Image: Portraits

GPT Image 2.5 Sunburst1,461.22 points

GPT Image 2.5 Flare1,439.29 points

GPT Image 21,426.27 points

MAI-Image-2.61,341.91 points

Grok Imagine Image 2.01,332.04 points

Reve 2.11,316.59 points

Muse Image1,289.88 points

Seedream 5.0 Pro1,286.94 points

GPT Image 1.51,258.96 points

Qwen-Image-2.11,256.2 points

Nano Banana Pro1,255.64 points

Nano Banana 21,254.9 points

MAI-Image-2.51,254.78 points

Nano Banana 2 Lite1,244.1 points

FLUX.2 Max1,156.6 points

FLUX.2 Pro1,144.32 points

FLUX.2 Flex1,143.85 points

Krea 2 Large1,123.58 points

HiDream-O1-Image1,105.31 points

FLUX.2 Klein 9B1,091.81 points

Stable Diffusion 3.5 Large941.86 points

higher is betterAs of November 6, 2026Scale: 0 points - 1,461.22 points

Source: LMArena

gradually.ai

Methodology and context

Metric: arena-score
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 21
Sample: 11,781
Methodology and context: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source: LMArena

LMArena Text to Image: Text rendering

GPT Image 2.5 Sunburst1,461.65 points

GPT Image 2.5 Flare1,431.87 points

GPT Image 21,424.21 points

MAI-Image-2.61,365.86 points

Grok Imagine Image 2.01,349.22 points

Reve 2.11,339.46 points

Nano Banana 21,294.44 points

Muse Image1,293.25 points

MAI-Image-2.51,280.73 points

Nano Banana 2 Lite1,273.63 points

Nano Banana Pro1,270.33 points

Seedream 5.0 Pro1,254.93 points

GPT Image 1.51,251.47 points

Qwen-Image-2.11,250.12 points

FLUX.2 Flex1,171.57 points

FLUX.2 Max1,165.6 points

FLUX.2 Pro1,154.35 points

HiDream-O1-Image1,134.31 points

Krea 2 Large1,116.42 points

FLUX.2 Klein 9B1,057.76 points

Stable Diffusion 3.5 Large921.07 points

higher is betterAs of November 6, 2026Scale: 0 points - 1,461.65 points

Source: LMArena

gradually.ai

Methodology and context

Metric: arena-score
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 21
Sample: 20,837
Methodology and context: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source: LMArena

Image editing

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.

GPT Image 2.5 Sunburst1,199.46 Elo

GPT Image 2.5 Flare1,182.04 Elo

MAI-Image-2.61,149.25 Elo

MAI-Image-2.6 Flash1,141.24 Elo

GPT Image 21,135.9 Elo

Seedream 5.0 Pro1,129.43 Elo

Muse Image1,128.02 Elo

Nano Banana 21,125.1 Elo

MAI-Image-2.51,124.35 Elo

GPT Image 1.51,116.41 Elo

Grok Imagine Image 2.01,115.93 Elo

Nano Banana Pro1,108.99 Elo

Grok Imagine Image Quality1,097.22 Elo

MAI-Image-2.5 Flash1,089.17 Elo

Qwen-Image-2.11,080.38 Elo

Ideogram 4.51,066.78 Elo

HunyuanImage 3.0 Instruct1,065.36 Elo

Qwen Image 2.0 Pro1,056.78 Elo

Nano Banana 2 Lite1,053.63 Elo

FLUX.2 Klein 9B1,012.79 Elo

FLUX.2 Pro1,011.07 Elo

FLUX.2 Max998.84 Elo

FLUX.2 Flex995.52 Elo

HiDream-O1-Image953.1 Elo

higher is betterAs of November 6, 2026Scale: 0 Elo - 1,199.46 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: 24
Methodology and context: AA appearances: 13849. This count does not establish an independent sample. Lower rank: 9. Upper rank: 16. Source model ID: 3f8a2d6b-9c4e-4f1a-b7d2-8e5a0c3d71f9. Evidence capture ID: aa-image-image-editing-v2.1-2026-11-06-0a4f79bb0933. Source payload hash: 0a4f79bb0933c8ef52ce1327ef382eb3442dab864e254d37ecd3176d39727555

Source: Artificial Analysis

LMArena Image Editing: Overall

GPT Image 2.5 Sunburst1,517.3 points

GPT Image 2.5 Flare1,482.13 points

GPT Image 21,460.09 points

Grok Imagine Image 2.01,438.04 points

MAI-Image-2.61,426.31 points

MAI-Image-2.51,400.28 points

Muse Image1,399.39 points

Seedream 5.0 Pro1,390.94 points

Grok Imagine Image Quality1,390.73 points

Nano Banana Pro1,390.2 points

Nano Banana 21,386.55 points

Reve 2.11,373.54 points

GPT Image 1.51,367.95 points

Qwen-Image-2.11,363.71 points

Ideogram 4.51,346.64 points

Nano Banana 2 Lite1,312.98 points

HunyuanImage 3.0 Instruct1,301.9 points

FLUX.2 Max1,261.49 points

FLUX.2 Pro1,248.1 points

FLUX.2 Flex1,224.55 points

FLUX.2 Klein 9B1,224.37 points

higher is betterAs of November 6, 2026Scale: 0 points - 1,517.3 points

Source: LMArena

gradually.ai

Methodology and context

Metric: arena-score
Scale: No upper bound is documented. The chart spans the recorded values and zero.
Participants: 21
Sample: 190,011
Methodology and context: Evidence capture ID: lmarena-image_edit-5a7e6bb84a93f55df56b198be8e56e53755a216a-4db53937a0d3. Source payload hash: 4db53937a0d39e716e49e4c3fb1b88c68578a094c4339bb7c714dbedfd0a85d0

Source: LMArena

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.

Nano Banana 299.7 rank points

FLUX.2 Flex98.1 rank points

FLUX 3 Image96 rank points

Recraft V4.1 Utility95.6 rank points

Wan 2.7 Image94.1 rank points

SenseNova U1.5 Lite93.1 rank points

Kling Image 3.0 Omni92.8 rank points

Nano Banana Pro91.9 rank points

FLUX.2 Max91.6 rank points

Recraft V4.1 Flash88.5 rank points

Grok Imagine Image Quality88.2 rank points

Krea 2 Large86.3 rank points

Firefly Image Model 585.7 rank points

Muse Image85.7 rank points

FLUX 3 Image Klein84.4 rank points

Seedream 4.583.8 rank points

MAI-Image-2.582.2 rank points

Nano Banana 2 Lite82.2 rank points

Qwen Image 2.0 Pro81 rank points

Wan 2.7 Image Pro81 rank points

HunyuanImage 3.0 Instruct78.8 rank points

Krea 377.6 rank points

Seedream 5.0 Flash76.6 rank points

HiDream-O1-Image-1.575.7 rank points

Seedream 5.0 Lite75.7 rank points

Luma Uni-1.1 Max75.1 rank points

GPT Image 274.5 rank points

Grok Imagine Image 1.073.8 rank points

Nano Banana 2.172.3 rank points

Krea 2 Medium Turbo71.3 rank points

Recraft V4.1 Utility Pro71 rank points

Grok Imagine Image 2.070.7 rank points

MAI-Image-2.5 Flash70.1 rank points

Recraft V4.169.5 rank points

FLUX.2 Pro68.9 rank points

Bria FIBO 268.5 rank points

Nano Banana (Gemini 2.5 Flash Image)68.2 rank points

ChatGPT Image Latest (API-Alias)67.9 rank points

GPT Image 1 Mini67.9 rank points

Seedream 5.0 Pro67.9 rank points

Ideogram 4.067.3 rank points

Luma Uni-1.167 rank points

Seedream 4.064.8 rank points

GPT Image 2.5 Flare63.9 rank points

LLaDA-Image62 rank points

MAI-Image-2.662 rank points

GPT Image 1.559.5 rank points

FLUX.2 Klein 9B Preview55.8 rank points

Reve 2.153.9 rank points

GPT Image 2.5 Sunburst53.6 rank points

MAI-Image-2.6 Flash53.6 rank points

Qwen-Image-2.153.6 rank points

Kling Image 3.051.4 rank points

Runway Gen-4 Image51.4 rank points

MOSS-Image-Base50.8 rank points

Qwen Image 2.1 Turbo50.8 rank points

GLM-Image50.2 rank points

FLUX.2 Dev49.8 rank points

FLUX.1 Kontext [max]45.5 rank points

Hy Image 3.5 Preview45.2 rank points

FLUX1.1 [pro]42.1 rank points

LongCat-Image41.4 rank points

Wan 2.6 Text to Image41.1 rank points

Z-Image Turbo39.6 rank points

Nucleus-Image38.3 rank points

Qwen Image 2.1 Pro38.3 rank points

MiniMax Image-0137.4 rank points

Chroma1-HD36.4 rank points

Bria FIBO Lite35.5 rank points

Qwen Image 3.0 Pro35.5 rank points

Z-Image Base32.7 rank points

FLUX.2 Klein 9B32.4 rank points

Qwen Image 3.032.4 rank points

FLUX.2 Klein 4B31.8 rank points

Stable Diffusion 3 Medium30.8 rank points

Stable Diffusion 3.5 Large30.8 rank points

OmniGen229.6 rank points

Ideogram 4.528.3 rank points

Vidu Q2 Image26.8 rank points

Stable Diffusion 3.5 Large Turbo25.5 rank points

FLUX1.1 [pro] Ultra21.8 rank points

Stable Diffusion 3.5 Medium21.8 rank points

MOSS-Image-SFT21.5 rank points

Kandinsky-5.0-T2I-Lite-SFT21.2 rank points

Adobe Firefly Image 420.6 rank points

Sana-Sprint 1.6B 1024px20.6 rank points

FLUX.1 Kontext [pro]19 rank points

Lumina-Image 2.018.4 rank points

Boogu-Image-0.1-Turbo18.1 rank points

HiDream-O1-Image18.1 rank points

Adobe Firefly Image 4 Ultra16.2 rank points

Chroma1-Flash16.2 rank points

FLUX.1 [dev]16.2 rank points

Kandinsky-5.0-T2I-Lite-Pretrain16.2 rank points

Sana-Sprint 0.6B 1024px15.9 rank points

Midjourney V8.114.3 rank points

HunyuanImage 3.012.5 rank points

AuraFlow v0.311.2 rank points

BAGEL-7B-MoT11.2 rank points

Boogu-Image-0.1-Base9 rank points

SANA-1.5 4.8B 1024px8.1 rank points

LLaDA-Image-Turbo7.8 rank points

Adobe Firefly Image 36.9 rank points

Stable Diffusion XL 1.05 rank points

SDXL Turbo3.4 rank points

PixArt-Sigma-XL-2-1024-MS3.1 rank points

SANA-1.5 1.6B 1024px3.1 rank points

Stable Diffusion Turbo0 rank points

Grok Imagine Image Quality

Krea 2 Large

Firefly Image Model 5

MAI-Image-2.5

HunyuanImage 3.0 Instruct

Grok Imagine Image 1.0

Nano Banana 2.1

Krea 2 Medium Turbo

Recraft V4.1 Utility Pro

Grok Imagine Image 2.0

Nano Banana (Gemini 2.5 Flash Image)

ChatGPT Image Latest (API-Alias)

FLUX.2 Klein 9B Preview

Reve 2.1

GPT Image 2.5 Sunburst

Wan 2.6 Text to Image

Stable Diffusion 3 Medium

Stable Diffusion 3.5 Large

OmniGen2

Ideogram 4.5

Vidu Q2 Image

Stable Diffusion 3.5 Large Turbo

FLUX1.1 [pro] Ultra

Stable Diffusion 3.5 Medium

MOSS-Image-SFT

Kandinsky-5.0-T2I-Lite-SFT

Adobe Firefly Image 4

Sana-Sprint 1.6B 1024px

FLUX.1 Kontext [pro]

Lumina-Image 2.0

Boogu-Image-0.1-Turbo

HiDream-O1-Image

Adobe Firefly Image 4 Ultra

Chroma1-Flash

FLUX.1 [dev]

Kandinsky-5.0-T2I-Lite-Pretrain

Sana-Sprint 0.6B 1024px

Adobe Firefly Image 3

Stable Diffusion XL 1.0

SDXL Turbo

PixArt-Sigma-XL-2-1024-MS

SANA-1.5 1.6B 1024px

Stable Diffusion Turbo

higher is betterAs of November 3, 2026Scale: 0 rank points - 100 rank points

Source: Gradually-Bildtest

gradually.ai

Methodology and context

Metric: rank-score-100
Scale: 0-100
Participants: 108
Sample: 1
Methodology and context: One archived image per model, using the first output of the accepted test attempt. Technical repeats are disclosed for the affected image. All 108 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: 1120.06
Score: 1120.06
Metric: elo-rating
Unit: Elo
Benchmark version: v2.1
Direction: Higher is better
Source rank: 11
Confidence interval: 11
Lower confidence bound: 1109.06
Upper confidence bound: 1131.06
Source type: independent-evaluator
Evaluator: Artificial Analysis
Retrieved at: 2026-11-06
Methodology: AA appearances: 14060. This count does not establish an independent sample. Lower rank: 8. Upper rank: 15. Source model ID: 3f8a2d6b-9c4e-4f1a-b7d2-8e5a0c3d71f9. Evidence capture ID: aa-image-text-to-image-v2.1-2026-11-06-e430d57b5d0b. Source payload hash: e430d57b5d0bf260d6feb60bb7d8fa32da8d7961ec2bd9a9cae58fb015613cf0
USD price per 1,000 tokens: 48.128

Source Artificial Analysis

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

Source value: 1124.35
Score: 1124.35
Metric: elo-rating
Unit: Elo
Benchmark version: v2.1
Direction: Higher is better
Source rank: 12
Confidence interval: 12
Lower confidence bound: 1112.35
Upper confidence bound: 1136.35
Source type: independent-evaluator
Evaluator: Artificial Analysis
Retrieved at: 2026-11-06
Methodology: AA appearances: 13849. This count does not establish an independent sample. Lower rank: 9. Upper rank: 16. Source model ID: 3f8a2d6b-9c4e-4f1a-b7d2-8e5a0c3d71f9. Evidence capture ID: aa-image-image-editing-v2.1-2026-11-06-0a4f79bb0933. Source payload hash: 0a4f79bb0933c8ef52ce1327ef382eb3442dab864e254d37ecd3176d39727555
USD price per 1,000 tokens: 48.128

Source Artificial Analysis

LMArena Text to Image: 3D modeling (Show measurement, test conditions, and source)

Source value: 1251.4749998591262
Score: 1251.4749998591262
Metric: arena-score
Unit: score
Direction: Higher is better
Source rank: 12
Sample: 5681
Confidence interval: 8.792342
Source type: official-leaderboard
Evaluator: LMArena
Retrieved at: 2026-11-06
Methodology: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source LMArena

LMArena Text to Image: Art (Show measurement, test conditions, and source)

Source value: 1276.7593080459058
Score: 1276.7593080459058
Metric: arena-score
Unit: score
Direction: Higher is better
Source rank: 10
Sample: 9115
Confidence interval: 7.5566
Source type: official-leaderboard
Evaluator: LMArena
Retrieved at: 2026-11-06
Methodology: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source LMArena

LMArena Text to Image: Cartoon (Show measurement, test conditions, and source)

Source value: 1268.645996588385
Score: 1268.645996588385
Metric: arena-score
Unit: score
Direction: Higher is better
Source rank: 14
Sample: 24138
Confidence interval: 5.827395
Source type: official-leaderboard
Evaluator: LMArena
Retrieved at: 2026-11-06
Methodology: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source LMArena

LMArena Text to Image: Commercial design (Show measurement, test conditions, and source)

Source value: 1259.3595988476006
Score: 1259.3595988476006
Metric: arena-score
Unit: score
Direction: Higher is better
Source rank: 13
Sample: 22289
Confidence interval: 5.862582
Source type: official-leaderboard
Evaluator: LMArena
Retrieved at: 2026-11-06
Methodology: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source LMArena

LMArena Text to Image: Overall (Show measurement, test conditions, and source)

Source value: 1252.9980152973003
Score: 1252.9980152973003
Metric: arena-score
Unit: score
Direction: Higher is better
Source rank: 15
Sample: 61222
Confidence interval: 4.176623
Source type: official-leaderboard
Evaluator: LMArena
Retrieved at: 2026-11-06
Methodology: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source LMArena

LMArena Text to Image: Photorealism (Show measurement, test conditions, and source)

Source value: 1250.667020609761
Score: 1250.667020609761
Metric: arena-score
Unit: score
Direction: Higher is better
Source rank: 18
Sample: 23657
Confidence interval: 5.822539
Source type: official-leaderboard
Evaluator: LMArena
Retrieved at: 2026-11-06
Methodology: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source LMArena

LMArena Text to Image: Portraits (Show measurement, test conditions, and source)

Source value: 1254.784489992374
Score: 1254.784489992374
Metric: arena-score
Unit: score
Direction: Higher is better
Source rank: 19
Sample: 11781
Confidence interval: 6.93076
Source type: official-leaderboard
Evaluator: LMArena
Retrieved at: 2026-11-06
Methodology: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source LMArena

LMArena Text to Image: Text rendering (Show measurement, test conditions, and source)

Source value: 1280.7272478853972
Score: 1280.7272478853972
Metric: arena-score
Unit: score
Direction: Higher is better
Source rank: 15
Sample: 20837
Confidence interval: 6.063817
Source type: official-leaderboard
Evaluator: LMArena
Retrieved at: 2026-11-06
Methodology: Evidence capture ID: lmarena-text_to_image-5a7e6bb84a93f55df56b198be8e56e53755a216a-43df833a690f. Source payload hash: 43df833a690f9454a344fb4d803c0c2ec845d592681e7f95e9ac50f1c0b25d4a

Source LMArena

LMArena Image Editing: Overall (Show measurement, test conditions, and source)

Source value: 1400.2754917238008
Score: 1400.2754917238008
Metric: arena-score
Unit: score
Direction: Higher is better
Source rank: 12
Sample: 190011
Confidence interval: 3.687129
Source type: official-leaderboard
Evaluator: LMArena
Retrieved at: 2026-11-06
Methodology: Evidence capture ID: lmarena-image_edit-5a7e6bb84a93f55df56b198be8e56e53755a216a-4db53937a0d3. Source payload hash: 4db53937a0d39e716e49e4c3fb1b88c68578a094c4339bb7c714dbedfd0a85d0

Source LMArena

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

Source value: 82.2
Score: 82.2
Metric: rank-score-100
Unit: rank score
Direction: Higher is better
Source rank: 17
Sample: 1
Source type: gradually-sample-test
Evaluator: gemini-3.1-pro-preview
Retrieved at: 2026-11-03
Methodology: One archived image per model, using the first output of the accepted test attempt. Technical repeats are disclosed for the affected image. All 108 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: 20
Individual ranks: 17, 26, 17
Criterion scores: headline 1.33, subline 1.33, text-control 1.33, visual-brief 1.33, layout 1.33

Source Gradually-Bildtest

Page 1 of 5

Popularity compared

How often the model is searched for, used and mentioned, compared with every model of its category.

Popularity index

The 100 most popular models of this category, 0 to 100, higher is more popular

20 of 100 models

Stable Diffusion XL 1.072

Anima Base v1.068

SDXL Turbo63

Stable Diffusion Turbo63

PixArt-Sigma-XL-2-1024-MS44

Janus-Pro-7B44

HunyuanImage 3.0 Instruct44

MOSS-Image-Base44

MAI-Image-2.543

Stable Diffusion XL 1.0

Anima Base v1.0

SDXL Turbo

Stable Diffusion Turbo

PixArt-Sigma-XL-2-1024-MS

Janus-Pro-7B

HunyuanImage 3.0 Instruct

MOSS-Image-Base

MAI-Image-2.5

As of November 2026 · Source: gradually.ai popularity index

Over time

Weekly values from October 18, 2026 to November 5, 2026

MAI-Image-2.5HiDream-O1-ImageHunyuanImage 3.0 InstructFLUX.2 Klein 9BMAI-Image-2.6

MAI-Image-2.5 compared

Compare with another model

First image modelSecond image model

The most popular comparisons

Show all 108 comparisons

All specifications

Architecture

Parameters: 20B; Source: Parameters

Input and output

Maximum output: 1,048,576 pixels; Source: Maximum output

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking: Unknown; no source
Watermarking technique: Not known; no source

Licence and access

Weights: Proprietary; Source: Weights
Access: API, Web; Source: Access
Lifecycle: Preview in Microsoft Foundry; Source: Lifecycle

Prices

PricesSubscriptions

Representative price: $0.048128 per imageSource
OpenRouter (microsoft/mai-image-2.5): $0.06 per image (estimated)Source

No Microsoft AI subscription includes this model. Through other providers from €19.04 a month (Magnific Premium, with limited credits). 6 plans in Germany.

AllIndividuals4Teams1Enterprise1

Plan	Monthly price	Note
Magnific Premium	€19.04	with limited credit
Magnific Premium+	€42.84	with limited credit
Magnific Business	€70.21	with limited credit, for teams, per seat
Magnific Pro Starter	€104.72	with limited credit
Magnific Pro	€273.70	with limited credit
Magnific Enterprise	n/a	for enterprises

All plans in the AI subscription comparison

Frequently asked questions about MAI-Image-2.5

How good is MAI-Image-2.5?

Rank 9 of 21 in the arena rating. Rank 26 of 108 in our own test.

Arena rating from Nov 6, 2026 · GraduallyBench from Nov 3, 2026

Arena ratingfrom user votes · higher is better

Reve 2.11,300

Muse Image1,268

Nano Banana 21,257

MAI-Image-2.5Rank 9 of 211,253

Seedream 5.0 Pro1,251

Nano Banana 2 Lite1,249

Nano Banana Pro1,246

GraduallyBenchour own rating, 0 to 100 · higher is better

Qwen Image 2.1 Turbo69

MAI-Image-2.5 Flash69

ChatGPT Image Latest (API-Alias)69

MAI-Image-2.5Rank 26 of 10869

FLUX.2 Max69

MAI-Image-2.668

GPT Image 1.568

What is MAI-Image-2.5 good at?

Best at Typography and layout (82 out of 100), weakest at Character and detail (55).

GraduallyBench, one image per task

Typography and layoutLegible text, hierarchy, and composition82out of 100Rank 17 of 108 in this task
Product photographyMaterials, reflections, and fine details70out of 100Rank 34 of 108 in this task
Character and detailAnatomy, clothing, and spatial consistency55out of 100Rank 50 of 108 in this task
InfographicNumbers, labels, and visual organization69out of 100Rank 38 of 108 in this task

How much does an image with MAI-Image-2.5 cost?

About $0.048 per image. That is fairly cheap: a typical image model costs $0.06.

Price per imageDollars per image · lower is better

Recraft V4.1 Utility$0.035

MAI-Image-2.6$0.039

HiDream-O1-Image$0.04

MAI-Image-2.5Rank 12 of 28$0.048

Grok Imagine Image Quality$0.05

Krea 2 Large$0.06

FLUX.2 Flex$0.06

Where can I use MAI-Image-2.5?

Not in a Microsoft AI subscription. Including other providers that makes 6 plans in Germany.

Subscription comparison, prices in Germany incl. VAT

At Microsoft AI

No Microsoft AI subscription includes this model.

Through other providers

6 plans from 1 provider, from €19.04 a month (Magnific Premium), with limited credits.

Sources and data date

Show all 25 sources

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

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

Without vendor claims and historical or disputed results; those stay labelled in the single values.

Type	Evidence and data date
Research status	Research date October 16, 2026. 13 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not published	immutable-serving-weights Note ImageBench names route fal/microsoft/mai-image-2.5; the CSV and README do not establish the exact checkpoint or immutable served weights.
Not published	generation-configuration-and-date Note The pinned aggregate CSV and README do not give this row's generation parameters, output resolution, or model-specific execution date.
Not published	category-and-eps-denominators Note Only the 192-prompt overall and pass-rate population is explicit. The aggregate sources provide no metric-specific N for EPS or category rates, so none is inferred from the total image count.
Not published	pixazo-per-model-sample-count-and-uncertainty Original wording The Pixazo page gives 10 prompt groups and 1,050 pairwise comparisons over 21 models but no per-model Elo denominator or confidence interval; the aggregate comparison count is not substituted for either.
Not published	pixazo-serving-identity-and-run-date Original wording Pixazo lists MAI-Image 2.5 (Microsoft) and a September 2026 leaderboard update, but not an immutable model-serving revision, generation settings, exact prompt set, or model-specific execution date.
Not published	exact-serving-revision Note The evaluator shows the exact product row but does not disclose an API route or serving checkpoint revision.
Unresolved	model-card sample size and topic-specific settings Note The card does not report the MAI-Image-2.5 evaluation N, topic prompt set, or a confidence interval for its published Elo column.
Additional source	Microsoft AI (retrieved October 19, 2026) · Microsoft MAI-Image-2.5 model card · Editorial description reviewed October 2, 2026
Additional source	Image pricing (retrieved October 17, 2026; October 28, 2026; November 3, 2026; November 5, 2026; November 6, 2026)
Additional source	Lifecycle data · Microsoft Foundry MAI image models documentation · Editorial description reviewed October 2, 2026
Additional source	OpenRouter
Additional source	Artificial Analysis (retrieved October 17, 2026; October 28, 2026; November 3, 2026; November 5, 2026; November 6, 2026)
Additional source	LMArena (retrieved November 6, 2026)1 2
Additional source	Gradually-Bildtest (retrieved November 3, 2026)
Additional source	ImageBench (retrieved October 16, 2026)
Additional source	Pixazo AI Image Generation Leaderboard 2026 (retrieved October 17, 2026)
Additional source	Massing Labs ArchViz Arena (retrieved October 17, 2026)
Additional source	twchrischang MAI-image-benchmark run summary (retrieved October 20, 2026)1 2
Additional source	Rapidata text-to-image human feedback benchmark (retrieved October 23, 2026)
Additional source	Rapidata benchmark.ai General image benchmark (retrieved October 28, 2026)
Additional source	Model metadata (retrieved October 16, 2026)
Additional source	Model metadata (retrieved October 17, 2026)
Additional source	Model metadata (retrieved October 28, 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
Additional source	Microsoft AI: Introducing MAI-Image-2.5 (2026-09-19)