Kimi K3.1 vs Opus 5.5 vs GPT-6: same cola geyser HTML prompt
Three models each built a single-HTML mint-in-cola geyser animation from the same prompt, compared on height, cost and time, with Opus 5.5 producing the biggest show.
Kimi K3.1 vs Claude Opus 5.5 vs GPT-6 Astra. same prompt: drop mints into a cola bottle, single html file. > Opus 5.5: 5.6 m geyser. $2.94, 16 min > Kimi K3.1: 4.4 m. $0.71, 5 min > GPT-6 Astra: 2.2 m and a cartoon "SPLOOSH!". $1.88, 10 min now per dollar: > Kimi: 6.2 m of geyser per $1 > Opus: 1.9 > Astra: 1.2 wait. nobody gave them the physics. each model picked its own pressure, its own foam, its own idea of how angry a cola bottle gets. so the height isn't accuracy. it's how much each model wanted the moment to land. Kimi got the most geyser per dollar. Opus built the biggest show. cheap sketch or the shot people replay. you're paying for one of those.
This post quotes another post: x.com/ai_screening/status/2113688082474221830