GPT-5.6 Terra in the animation coding benchmark
GPT-5.6 Terra by OpenAI takes the PomBench animation coding challenge: create a twelve-second scene of a miniature Pomeranian jumping over a puddle, entering a doghouse, retrieving one bone and turning toward the viewer. Watch the recorded result above, then compare its movement and scene coherence with other AI models given the same initial brief.
The model generates self-contained browser code rather than a video file. An independent offline browser records that code at 960 × 540 pixels and 30 frames per second. Look for a grounded running gait, a readable jump and landing, a reflection inside the puddle, correct doorway occlusion and a bone that stays in the dog's mouth during the final turn.
For this run, 8,121 output tokens were recorded, including provider-reported reasoning; the known API cost is $0.1183. This result includes a model-authored execution repair. The final response is shown, and the reported usage includes every completed response. This is a sample of animation programming, not a general intelligence ranking or an automatically awarded visual score.