Compare up to four models on the same twelve-second Pomeranian scene. The player synchronizes playback, seeking and speed, so you can inspect the same moment across models.
Every model writes its own native browser animation. Compare running, the puddle jump and reflection, doorway occlusion, bone retrieval and the final turn toward the camera. Output tokens include reasoning; known costs include completed responses and documented repairs.
Looking for the best LLM for coding? Watch the same task side by side
PomBench does not crown a single winner. It shows you how each LLM handles the same practical coding brief, a complete browser animation, and lets you judge the result yourself. Pick two to four models, press play, and compare movement, scene coherence, output tokens and known API cost on one timeline.
It depends on the task. PomBench tests one demanding task: writing a self-contained, twelve-second browser animation from a shared prompt. Watch the recordings to see which models produce coherent movement, then weigh output tokens and known API cost. It is one sample of programming ability, not an overall ranking.
How do I compare AI models on PomBench?
Choose up to four models in the player above. Playback, seeking and speed are synchronized, so the same moment of the scene is shown for every model.
Is every model given the same prompt?
Yes. Every model receives the same initial prompt and output allowance (32,768 tokens including reasoning, subject to lower provider ceilings). One execution repair is available for empty, truncated or unexecutable code, and no organizer edits model code. Read the methodology.
What do output tokens and known cost mean?
Output tokens are provider-reported and include reasoning once. Known cost uses recorded input and output at dated API rates, or the provider-reported charge for OpenRouter results. Missing prices are shown as unavailable, not zero.
