A Juejin AI blogger published a hands-on test of Kimi K3, describing it as clearly stronger than the previous K2.7 on multi-modal front-end and 3D tasks, with a clean plan-then-verify workflow, but uneven across tasks. In the author's tests K3 handled a 3D garage page and a Super Mario recreation well, while it produced a flawed 3D billiards game and a less rich "cyberpunk Along the River During the Qingming Festival" scene than Fable 5. The author reports that access to K3 requires a 99-yuan-or-higher plan and that 1M context requires the 199-yuan tier, and that roughly five test pages exhausted a week's quota in under a day, with the first example alone using about 37% of the weekly allowance.