Gemini 4 Krypton has not been released
The post shows a four-model comparison of horse illustrations and claims they were made with the same prompt. This is an unverified community showcase, not a basis for ranking the models.
Tasks this can help assess: See the four horse illustrations and the labels shown in the author's post.
Claims this cannot support: Inferring overall model capability, consistency, or ranking, or confirming that all four generations used identical conditions.
Model versions shown: Gemini 4 Krypton, Claude Fable 5.2, Claude Opus 5.5, and DeepSeek V4.1 Flash. Gemini 4 Krypton is the display name used in the post; its actual model ID has not been confirmed, and it has not launched yet.
Test environment or client: Not specified.
Reasoning level and parameters: Not specified.
The author says the four models used the "same prompt," but did not publish the prompt, invocation method, model IDs, parameters, or generation process. This review examined only the visible preview frames of the embedded video in the post. The video was not played, and no independent retest was conducted.
The preview labels four panels with the names of the four models and shows a horse illustration in each. DeepSeek V4.1 Flash's horse is made of branch-like elements; Gemini 4 Krypton shows a running horse against a natural sunset background; Claude Opus 5.5 and Claude Fable 5.2 each show a horse illustration as well. The four images visibly differ in style and form. The post does not say which image is better, provide scores, or give an explicit ranking.
The original post says "Same prompt. Four models" and labels the four models by their displayed names.
The visible preview is a four-panel comparison of horse illustrations; the post lists the video as 16 seconds long.
The sample consists only of this one set of images shown in the post. The prompt and other experimental conditions were not disclosed.
This is a community comparison showcase, not a reproducible independent test. The four names in the image are labels supplied by the author and do not establish the actual identities of the models used. "Same prompt" is also only a claim in the original post. No scores or conclusions that would support a winner are visible, so the differences between the illustrations cannot be used to rank model capability.
The original conditions cannot be reproduced because the post does not disclose the prompt, model IDs, client, parameters, or generation records. This note records only what is visible in the post text and preview frames.