Head-to-head comparison

GPT Image 1.5 vs. MAI-Image-2.5

Compare shared capability tests, arena scores with documented retrieval dates, task-specific image pricing, and verified specifications. Research benchmarks, preference arenas, and our n = 1 samples remain separate.

shared result series: 15
shared sample prompts: 4
latest arena retrieval: 2026-09-22

First image modelSecond image model

Specifications and cost

General specifications and task-specific pricing

A model can have one price for generating new images and another for editing. Prices therefore remain tied to the corresponding arena configuration.

Specifications and pricing for GPT Image 1.5 and MAI-Image-2.5
Attribute	GPT Image 1.5	MAI-Image-2.5
Provider	OpenAI	Microsoft AI
Released	December 2025	June 2026
Access	API, Web app	API, Web app
Lifecycle	Marked as deprecated by OpenAI; removed from the API on December 1, 2026 Status source	Preview in Microsoft Foundry Status source
Weights	Proprietary	Proprietary
Parameters	Not published	20 billion
Maximum output	1,536 × 1,024 pixels	1,048,576 pixels
Arena configuration	High	No separate configuration named
Text-to-image price	$133 per 1,000 images	$48.1 per 1,000 images
Editing price	$133 per 1,000 images	$48.1 per 1,000 images
Technical source	Model sourceVerified 2026-08-15	Model sourceVerified 2026-08-15

GPT Image 1.5MAI-Image-2.5

Text to Image ArenaUSD per 1,000 images at 1,024 × 1,024 pixels

GPT Image 1.5$133

MAI-Image-2.5$48.1

Image Editing ArenaUSD per 1,000 images at 1,024 × 1,024 pixels

GPT Image 1.5$133

MAI-Image-2.5$48.1

Key to the marks:better valuebetter, but not conclusivetie

Prompt adherence and specialist tests

Shared capability benchmarks

Both checkpoints completed the same four Gradually sample tasks. For each task, 30 archived first outputs are ranked in three differently ordered, model-blind passes. The published 0-100 value normalizes the mean rank. It must still be read as an n = 1 sample per model. No exact shared result from the integrated research benchmarks is available for this pair.

GPT Image 1.5MAI-Image-2.5

Gradually sample test: Typography and layoutMean of three blind ranks across all 30 first outputs, n = 1 each

GPT Image 1.529.9 / 100

MAI-Image-2.582.8 / 100

Gradually sample test: Product photographyMean of three blind ranks across all 30 first outputs, n = 1 each

GPT Image 1.543.7 / 100

MAI-Image-2.579.3 / 100

Gradually sample test: Character and detailMean of three blind ranks across all 30 first outputs, n = 1 each

GPT Image 1.577 / 100

MAI-Image-2.520.7 / 100

Gradually sample test: InfographicMean of three blind ranks across all 30 first outputs, n = 1 each

GPT Image 1.542.5 / 100

MAI-Image-2.564.4 / 100

Key to the marks:better valuebetter, but not conclusivetie

Capability benchmark scores for GPT Image 1.5 and MAI-Image-2.5
Benchmark	GPT Image 1.5	MAI-Image-2.5
Gradually sample test: Typography and layout	29.9 / 100 Mean rank: 21.33 (23 / 22 / 19) Five rubric scores Headline: 1.67 / 2 Subline: 1.67 / 2 Text control: 1.67 / 2 Visual brief: 1.33 / 2 Layout: 1 / 2	Higher blind-rank score82.8 / 100 Mean rank: 6 (3 / 7 / 8) Five rubric scores Headline: 1.67 / 2 Subline: 1.67 / 2 Text control: 1.67 / 2 Visual brief: 1.67 / 2 Layout: 1.67 / 2
Gradually sample test: Product photography	43.7 / 100 Mean rank: 17.33 (22 / 8 / 22) Five rubric scores Subject: 1.67 / 2 Time: 1 / 2 Watch details: 1.67 / 2 Photography: 1.67 / 2 Exclusions: 1.67 / 2	Higher blind-rank score79.3 / 100 Mean rank: 7 (6 / 12 / 3) Five rubric scores Subject: 1.67 / 2 Time: 1.67 / 2 Watch details: 1.67 / 2 Photography: 1.67 / 2 Exclusions: 1.67 / 2
Gradually sample test: Character and detail	Higher blind-rank score77 / 100 Mean rank: 7.67 (6 / 5 / 12) Five rubric scores Scene: 1 / 2 Wardrobe: 1 / 2 Anatomy: 1 / 2 Props: 1 / 2 Environment: 1 / 2	20.7 / 100 Mean rank: 24 (24 / 21 / 27) Five rubric scores Scene: 1 / 2 Wardrobe: 1 / 2 Anatomy: 1 / 2 Props: 1 / 2 Environment: 1 / 2
Gradually sample test: Infographic	42.5 / 100 Mean rank: 17.67 (20 / 14 / 19) Five rubric scores Title: 1.67 / 2 Sequence: 1.67 / 2 Labels: 1 / 2 Diagram: 1.67 / 2 Clarity: 1 / 2	Higher blind-rank score64.4 / 100 Mean rank: 11.33 (11 / 12 / 11) Five rubric scores Title: 1.67 / 2 Sequence: 1.33 / 2 Labels: 1 / 2 Diagram: 1.67 / 2 Clarity: 1.33 / 2
Key to the marks:better valuebetter, but not conclusivetie

Independent preference tests

Shared arena benchmarks

Artificial Analysis and Arena derive scores from blind comparisons. Raters see outputs for the same prompt and choose the image they prefer. Separate series cover text rendering, photorealism, portraits, art, commercial design, and image editing. Confidence intervals indicate how certain a measured lead is.

11 shared result series

GPT Image 1.5MAI-Image-2.5

Text to Image ArenaBlind preference from the same prompt, higher is better

GPT Image 1.51,102 Elo

MAI-Image-2.51,103 Elo

Image Editing ArenaBlind preference from the same prompt, higher is better

GPT Image 1.51,103 Elo

MAI-Image-2.51,113 Elo

LMArena Text to Image: 3D modelingBlind preference from the same prompt, higher is better

GPT Image 1.51,214.044

MAI-Image-2.51,248.552

LMArena Text to Image: ArtBlind preference from the same prompt, higher is better

GPT Image 1.51,224.786

MAI-Image-2.51,277.455

LMArena Text to Image: CartoonBlind preference from the same prompt, higher is better

GPT Image 1.51,245.019

MAI-Image-2.51,269.579

LMArena Text to Image: Commercial designBlind preference from the same prompt, higher is better

GPT Image 1.51,240.235

MAI-Image-2.51,260.92

LMArena Text to Image: OverallBlind preference from the same prompt, higher is better

GPT Image 1.51,239.16

MAI-Image-2.51,254.072

LMArena Text to Image: PhotorealismBlind preference from the same prompt, higher is better

GPT Image 1.51,249.33

MAI-Image-2.51,250.943

LMArena Text to Image: PortraitsBlind preference from the same prompt, higher is better

GPT Image 1.51,261.708

MAI-Image-2.51,257.755

LMArena Text to Image: Text renderingBlind preference from the same prompt, higher is better

GPT Image 1.51,253.437

MAI-Image-2.51,282.526

LMArena Image Editing: OverallBlind preference from the same prompt, higher is better

GPT Image 1.51,370.19

MAI-Image-2.51,400.21

Key to the marks:better valuebetter, but not conclusivetie

Arena benchmark scores for GPT Image 1.5 and MAI-Image-2.5
Benchmark	GPT Image 1.5	MAI-Image-2.5	Source
Text to Image Arena	1,102 Elo Rank 9, 95% confidence interval: ±9, 14,748 ratings	No clear lead1,103 Elo Rank 8, 95% confidence interval: ±9, 14,914 ratings	Artificial AnalysisRetrieved 2026-09-22
Image Editing Arena	1,103 Elo Rank 12, 95% confidence interval: ±9, 12,789 ratings	No clear lead1,113 Elo Rank 7, 95% confidence interval: ±9, 12,485 ratings	Artificial AnalysisRetrieved 2026-09-22
LMArena Text to Image: 3D modeling	1,214.044 Rank 15, 95% confidence interval: ±6.6, 15,526 ratings	Higher score1,248.552 Rank 11, 95% confidence interval: ±9.04, 5,230 ratings	LMArenaRetrieved 2026-09-19
LMArena Text to Image: Art	1,224.786 Rank 14, 95% confidence interval: ±5.9, 21,424 ratings	Higher score1,277.455 Rank 7, 95% confidence interval: ±7.76, 8,400 ratings	LMArenaRetrieved 2026-09-19
LMArena Text to Image: Cartoon	1,245.019 Rank 14, 95% confidence interval: ±4.72, 59,294 ratings	Higher score1,269.579 Rank 11, 95% confidence interval: ±5.93, 22,528 ratings	LMArenaRetrieved 2026-09-19
LMArena Text to Image: Commercial design	1,240.235 Rank 14, 95% confidence interval: ±4.6, 58,887 ratings	Higher score1,260.92 Rank 10, 95% confidence interval: ±5.97, 20,864 ratings	LMArenaRetrieved 2026-09-19
LMArena Text to Image: Overall	1,239.16 Rank 15, 95% confidence interval: ±3.24, 151,702 ratings	Higher score1,254.072 Rank 12, 95% confidence interval: ±4.28, 57,250 ratings	LMArenaRetrieved 2026-09-19
LMArena Text to Image: Photorealism	1,249.33 Rank 15, 95% confidence interval: ±4.7, 60,842 ratings	No clear lead1,250.943 Rank 14, 95% confidence interval: ±5.89, 21,971 ratings	LMArenaRetrieved 2026-09-19
LMArena Text to Image: Portraits	No clear lead1,261.708 Rank 12, 95% confidence interval: ±5.81, 29,731 ratings	1,257.755 Rank 13, 95% confidence interval: ±7.1, 10,873 ratings	LMArenaRetrieved 2026-09-19
LMArena Text to Image: Text rendering	1,253.437 Rank 16, 95% confidence interval: ±4.84, 54,527 ratings	Higher score1,282.526 Rank 11, 95% confidence interval: ±6.17, 19,611 ratings	LMArenaRetrieved 2026-09-19
LMArena Image Editing: Overall	1,370.19 Rank 15, 95% confidence interval: ±2.81, 566,510 ratings	Higher score1,400.21 Rank 7, 95% confidence interval: ±3.76, 174,744 ratings	LMArenaRetrieved 2026-09-19
Key to the marks:better valuebetter, but not conclusivetie

Visual comparison

Sample images

Our images complement preference data, but do not replace it. The prompt, request count, and selection rule match. Provider resolution, quality tier, randomness, and prompt processing remain visible limitations, so this is not a fully controlled laboratory test.

Largely standardized comparison protocol. Both models receive the same archived prompt, exactly one request, and the first result counts. Files are normalized to 1,024 × 1,024 pixels without cropping. Provider resolution, randomness, quality tier, and prompt processing can still differ. These images are a practical visual test, not a reconstruction of the arena configuration.

Character and detail

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.

GPT Image 1.5. Provider Together, 1024x1024, n=1, seed=provider random, provider default steps, model ID openai/gpt-image-1.5, generated Aug 14, 2026, 12:17 PM. Source
MAI-Image-2.5. Provider OpenRouter, aspect ratio 1:1, provider default 1K resolution, n=1, seed=provider random, actual cost $0.048508, model ID microsoft/mai-image-2.5, generated Aug 14, 2026, 1:26 PM. Source

Protocol gradually-image-comparison-v1, selection rule: first result.

Prompt SHA 256: 985a75b38642a9f0e5e0c0e0c67f1b7f67780d42bd2168fa24e9a7a7a94b0da6

GPT Image 1.5, image SHA 256: 0dff52ccd39400ac66ac8870c9ea937512ff1474688e70118688f420c38c2296

GPT Image 1.5, request SHA 256: 5199631ac6ff07633dd2fb6ab61c918c26e1e02f37e6131828072233ac04527d

MAI-Image-2.5, image SHA 256: ca1c58688ae4601975d7e1f7b58bb43a0a999e736f610d9f51cd11e06284bced

MAI-Image-2.5, request SHA 256: 40fc54cbcee64f9045c44e702e013b649a0d8bbd7f8e2390d839bc38ee183b90

Infographic

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.

GPT Image 1.5. Provider Together, 1024x1024, n=1, seed=provider random, provider default steps, model ID openai/gpt-image-1.5, generated Aug 14, 2026, 12:17 PM. Source
MAI-Image-2.5. Provider OpenRouter, aspect ratio 1:1, provider default 1K resolution, n=1, seed=provider random, actual cost $0.048538, model ID microsoft/mai-image-2.5, generated Aug 14, 2026, 1:26 PM. Source

Protocol gradually-image-comparison-v1, selection rule: first result.

Prompt SHA 256: ed65cb4e6a51e2336f8b149ad4cbb16e020f07d527ad94362e719d2dcf91d94f

GPT Image 1.5, image SHA 256: 1c9ff59c32621c8b48b002dacb0ae66487262280c7c0a5b3d6456c81b248ebe9

GPT Image 1.5, request SHA 256: 7f1efc9531f1e6bed7d61a11c637add9fd892001aea2bb2781f8d89c79a2eb03

MAI-Image-2.5, image SHA 256: eaf00a906231e8859f98ef5dc2cce7e8017797819aea453305a4ef0e970d4dbf

MAI-Image-2.5, request SHA 256: bbedc701e9933a8d6ed0fbb2a6fbb7214214fad9d32d9ae80a114a7254a0bcf1

Product photography

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.

GPT Image 1.5. Provider Together, 1024x1024, n=1, seed=provider random, provider default steps, model ID openai/gpt-image-1.5, generated Aug 14, 2026, 12:16 PM. Source
MAI-Image-2.5. Provider OpenRouter, aspect ratio 1:1, provider default 1K resolution, n=1, seed=provider random, actual cost $0.048513, model ID microsoft/mai-image-2.5, generated Aug 14, 2026, 1:25 PM. Source

Protocol gradually-image-comparison-v1, selection rule: first result.

Prompt SHA 256: c67fd71e91304458a2feef5c91efa8677260fab0c0099a29a2c52cb579f5b2f3

GPT Image 1.5, image SHA 256: 67cfa2f324447027ceb5f82e9446d40a1395b46ce425890a74606b4d97731cf6

GPT Image 1.5, request SHA 256: c850843a298631a4076c5d2686de8e9d7223534f2b8ca53505f437862718bef9

MAI-Image-2.5, image SHA 256: 67d4930b711fac61482231c4bc270cfcdb41c1b0daf653045781fb2126e237dd

MAI-Image-2.5, request SHA 256: 071e156eeb3e65e6744b7c6b675ffcb32b551dcfc1033a55e30e9561c0bd78a9

Typography and layout

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.

GPT Image 1.5. Provider Together, 1024x1024, n=1, seed=provider random, provider default steps, model ID openai/gpt-image-1.5, generated Aug 14, 2026, 1:59 PM. Source
MAI-Image-2.5. Provider OpenRouter, aspect ratio 1:1, provider default 1K resolution, n=1, seed=provider random, actual cost $0.048473, model ID microsoft/mai-image-2.5, generated Aug 14, 2026, 2:03 PM. Source

Protocol gradually-image-comparison-v1, selection rule: first result.

Prompt SHA 256: 2e012c3d14ee5dd02203a94f394ca80a5038cdebd6032b25a8b8a221f2255419

GPT Image 1.5, image SHA 256: 6a4cec86850fd371b65274198e2bf630b3cf36a096f9fb953d4bfe17c3222ebe

GPT Image 1.5, request SHA 256: 9c3a396aecfcaae3f8a25edbe5c8c44d0d33199d7ba29efda864d2b0221c3250

MAI-Image-2.5, image SHA 256: 4c8a60555b57b8c3b5ebaa131371d1bba9d31bfafe7aaa772554969f7c28cb08

MAI-Image-2.5, request SHA 256: f14577273bd6b0980f49e13988b01555bb55dc824317b201b5fab3d2ef72e666

How to read this comparison

An Elo score is not a verdict on creativity

A higher arena score means this specific configuration was preferred more often in blind comparisons.

Typography, character consistency, local execution, licensing, and price may matter more to your workload than overall visual appeal.

Prices cover 1,000 images at 1,024 × 1,024 pixels using settings documented by Artificial Analysis. Subscriptions and special rates are excluded.

More matchups

Compare GPT Image 1.5 and MAI-Image-2.5 with other important image models under the same evidence rules. Each page shows only exact shared measurements and samples.

View the complete image model comparison

GPT Image 1.5 vs. MAI-Image-2.5

Change image model selection

General specifications and task-specific pricing

Shared capability benchmarks

Shared arena benchmarks

Sample images

Character and detail

Infographic

Product photography

Typography and layout

An Elo score is not a verdict on creativity

Related image model comparisons