Pat Simmons gives Kimi K3, Fable 5 and GPT-5.6 Sol the same prompts across an interactive ocean world, a cinematic camera website, a 3D game, landing-page concepts and a brand-revival deck. The outputs reveal large differences by task rather than one universal winner.
Fable 5 delivered the most polished results on demanding 3D scenes, but its runs were several times more expensive. Kimi K3 frequently came close, showed strong persistence and cost much less, although camera placement, interaction logic and one-shot quality remained uneven.
On simpler design and knowledge-work tasks, GPT-5.6 Sol sometimes produced the preferred result quickly and cheaply. Pat Simmons therefore recommends routing work by complexity: reserve expensive frontier effort for the hardest builds and use lower-cost models for routine production.
Watch the original on YouTube