OpenAI
1306 Elo ±45
Rank #2 of 42 · median 126s per image · 12 refused
Not judged in the same call yet
Elo expectation: GPT Image 2 wins 78%
Reve
1087 Elo ±36
Rank #16 of 42 · median 53s per image

By category

CategoryGPT Image 2Reve 2.1Direct record
Anime & Cartoon Style 1344 1003 —
Architecture & Interiors 1531 1094 —
Complex Scenes 1443 1198 —
Ghibli style 1195 1006 —
Graphic Design 1208 1190 —
Hands & Anatomy 1128 963 —
Photorealistic People & Portraits 1349 1072 —
Surreal & Creative Prompts 1366 1081 —
Text in Images 1356 1155 —
Ultra Hard 1334 1151 —

By criterion

CriterionGPT Image 2Reve 2.1
Prompt adherence 1219 1077
Technical quality 1235 1082
Artistic merit 1213 982
Detail execution 1276 1082
Realism & coherence 1239 1078

Side by side

A magical girl character in a frilly dress, casting a spell (classic anime trope). - GPT Image 2
GPT Image 2 9 / 10 This is another leading image, with excellent trope recognition and highly polished character rendering. It falls just behind the top entry because its foreground hand and crowded effects are slightly less controlled.
A magical girl character in a frilly dress, casting a spell (classic anime trope). - Reve 2.1
Reve 2.1 6 / 10 Ranked 7th of 13, this has strong immediate appeal and complete prompt coverage. Unconvincing foreground-hand proportions and noisy garment construction substantially weaken its finish.
Photorealistic rendering of a contemporary glass skybridge between two skyscrapers, seen from inside with city views visible through glass walls and floor. Show accurate reflections and structural details. - GPT Image 2
GPT Image 2 9 / 10 This is among the two strongest images, especially for restrained photographic realism. Its clear destination and convincing glass behavior distinguish it from the more ambiguous mirrored corridors.
Photorealistic rendering of a contemporary glass skybridge between two skyscrapers, seen from inside with city views visible through glass walls and floor. Show accurate reflections and structural details. - Reve 2.1
Reve 2.1 6 / 10 Ranked 8th of 13 because it meets the main scene requirements but lacks convincing material separation and construction specificity. It provides better downward visibility than the opaque-floor examples without matching the leaders' clarity.
A school classroom of children and a teacher, each student engaged in a different activity. - GPT Image 2
GPT Image 2 9 / 10 Ranked first of 13, this provides the strongest combination of differentiated activities, believable interaction, and photographic execution. Minor task repetition prevents a perfect score, but it fulfills the central concept more convincingly than the other images.
A school classroom of children and a teacher, each student engaged in a different activity. - Reve 2.1
Reve 2.1 6 / 10 Ranked 8 of 13, this is polished and readable but substantially repeats the same student activity. Strong environmental detail does not fully compensate for limited differentiation and staged group dynamics.
A child and elder tending oversized vegetables in a garden. Reproduce the artistic style of Studio Ghibli, with their distinctive character proportions and detailed plant illustrations. - GPT Image 2
GPT Image 2 8 / 10 Ranks fifth, offering strong stylistic fidelity and believable teamwork. Its subdued contrast and crowded vegetation make it slightly less immediately compelling than the higher-ranked scenes.
A child and elder tending oversized vegetables in a garden. Reproduce the artistic style of Studio Ghibli, with their distinctive character proportions and detailed plant illustrations. - Reve 2.1
Reve 2.1 8 / 10 Ranked 4th of 13: the affectionate interaction, convincing contact, and strong natural textures distinguish this image. Small botanical simplifications and crowded peripheral foliage place it just below the leading compositions.
Abstract geometric logo for tech startup 'Quantum Leap', evoking connection & speed, vibrant blue/purple gradient, sharp angles. 'Quantum Leap' as text contained in logo. - GPT Image 2
GPT Image 2 9 / 10 Ranked 1st of 13. It offers the strongest combination of prompt fidelity, concept communication, readable integrated text, and disciplined visual impact.
Abstract geometric logo for tech startup 'Quantum Leap', evoking connection & speed, vibrant blue/purple gradient, sharp angles. 'Quantum Leap' as text contained in logo. - Reve 2.1
Reve 2.1 5 / 10 Ranked 10th of 13: readable typography prevents this from being unsuccessful outright. Nevertheless, the disjointed frame and poorly prioritized decoration make it noticeably weaker as a cohesive logo.
A realistic photo of a hand holding a red apple, with all fingers clearly visible. - GPT Image 2
GPT Image 2 9 / 10 Ranked 3rd of 13 for its exceptionally clear five-digit presentation and strong realism. It falls behind the top two mainly in visual impact, not in basic execution.
A realistic photo of a hand holding a red apple, with all fingers clearly visible. - Reve 2.1
Reve 2.1 5 / 10 Ranks 11th of 13 because it combines an incomplete finger display with somewhat artificial surface rendering. Its clean composition is not enough to overcome those task-relevant shortcomings.
A studio headshot of a young man with one blue eye and one green eye (heterochromia). - GPT Image 2
GPT Image 2 9 / 10 Ranked 2nd of 13, this is an excellent naturalistic rendering of the requested trait. The leading portrait has slightly greater lighting depth and visual character, but the difference is small.
A studio headshot of a young man with one blue eye and one green eye (heterochromia). - Reve 2.1
Reve 2.1 4 / 10 Ranked 13th of 13, this is a technically competent portrait that misses the prompt's central distinguishing requirement. The absence of clear blue-green heterochromia outweighs its otherwise credible photographic presentation.
A dessert cake that looks like a tiny planet, complete with miniature trees and mountains, photographed in a professional food photography style with tilt-shift focus effect. - GPT Image 2
GPT Image 2 8 / 10 Ranked 4th of 13 for rich miniature-world construction and strong photographic depth. A less decisive globe silhouette and slightly crowded decoration keep it below the leading three.
A dessert cake that looks like a tiny planet, complete with miniature trees and mountains, photographed in a professional food photography style with tilt-shift focus effect. - Reve 2.1
Reve 2.1 5 / 10 Ranked 11th of 13, this succeeds as a detailed landscape cake but misses the defining planetary appearance. That central prompt deviation outweighs its otherwise solid photographic execution.
A wrinkled white T-shirt with "Carpe Diem" in bold serif font and "Seize The Day" in smaller script below it, modeled on a mannequin, professional product photography with soft-box lighting. - GPT Image 2
GPT Image 2 9 / 10 Ranks second of 13 and is another excellent match to the complete brief. Its slightly busier studio framing places it just behind the cleaner, more concentrated presentation of image 3995.
A wrinkled white T-shirt with "Carpe Diem" in bold serif font and "Seize The Day" in smaller script below it, modeled on a mannequin, professional product photography with soft-box lighting. - Reve 2.1
Reve 2.1 6 / 10 Ranked 9th of 13, this is a clean photograph with accurate typography. Its failure to show substantial wrinkling makes it a weaker response to this particular prompt.
Photorealistic depiction of a man wearing a clearly visible black OpenAI-branded T-shirt. He is standing at the front of a university lecture hall, writing complex mathematics and AI-related equations across a large, dusty chalkboard filled with notation - GPT Image 2
GPT Image 2 9 / 10 Ranked 2nd of 13, this is an exceptionally convincing classroom photograph with strong equation generation. It falls slightly behind the winner in complete equation visibility and overall presentation clarity.
Photorealistic depiction of a man wearing a clearly visible black OpenAI-branded T-shirt. He is standing at the front of a university lecture hall, writing complex mathematics and AI-related equations across a large, dusty chalkboard filled with notation - Reve 2.1
Reve 2.1 6 / 10 The image succeeds as an attractive branded teaching portrait but only partially fulfills the requested activity. Incorrect equation rendering and the absence of visible writing keep it below the stronger lecture scenes.

More comparisons