Code2Video Bench: Benchmark tests LLM agents generating code that renders to video

Google, Kaggle, and HeyGen launch Code2Video Bench, a benchmark that tests whether an LLM agent can write code that a rendering engine turns into a final video.

Paige Bailey, a Google developer relations lead, promotes the benchmark on X. She says that given a prompt or visual narrative, the agent should generate code that a framework such as HyperFrames renders deterministically into the video.

HyperFrames says the benchmark was built with Google DeepMind and Kaggle to measure agentic video generation. It distinguishes this from SWE-bench-style code tasks.