GPT-6, Haiku 5.5 and DeepSeek take on the same slime-through-grate physics animation
The same prompt asked three models to simulate slime squeezing through a metal grate at different viscosities: GPT-6 Luna took 18 minutes for $0.22, Haiku 5.5 took 35 minutes for $1.67, and DeepSeek ran nearly 3 hours with a grate but no slime.
GPT-6 LUNA: $0.22 SLIME. DEEPSEEK: 3 HOURS OF ABSOLUTELY NOTHING. 😭 Three models. Same starting brief: make slime squeeze through a metal grate at different viscosities. GPT-6 Luna High · 18:34 · 84.6K tokens · $0.22 Haiku 5.5 High · 35:02 · 225K tokens · $1.67 DeepSeek V4.1 Flash High · 2:57:48 · 255.5K tokens · $0.76 Luna kept it thick and connected. Haiku turned it into a shower of green blobs. DeepSeek spent nearly 3 hours in its own Harness. The grate showed up. The slime didn’t. This http://aimlapi.com physics experiment got ridiculous.
This post quotes another post: x.com/noclipepe/status/2102572756353061250