So, I tested Gemini 4 argon. Used the same starting point as Claude. The video below compares the landing page made by claude vs gemini. It's a little unfair because the Claude version is a result of 18 detailed prompts telling Claude what changes to make. However it's still a valid comparison because Claude one shotted something that was very much usable.
The second video shows Gemini's brilliant redesign after I told it to make the site look like DaVinci Resolve. Note that I didn't tell Claude to make it look and feel like Resolve -- Claude asked an MCQ and I clicked "Gradnig Suite".
Gemini didn't ask questions. It just started working and made what it made. I don't think it works even as a starting point. It also made really silly mistakes like completely deleting image links.
I have no doubt in my mind that Gemini 4 Argon is a benchmaxxed model.
+7
모델 비교신규0:24눈 덮인 숲의 여우: 비교@karabolamani프롬프트 공개
모델 비교신규0:50베네치아 홍수 시뮬레이션: 비교@Fanfulladev
모델 비교신규0:16문명 시뮬레이션: 비교@remoteugur
모델 비교신규0:21방콕 사진을 3D로: 비교@Avenoxai
모델 비교신규0:15마리오네트: 비교@ClearTextAI
모델 비교신규0:18병 속의 배: 비교@vib3coded