Show HN: Jev vs. GPT-5.6 and Claude Haiku at Pong

Jev TypeSafe AI uses a typed-decision model to play Atari Pong, while Anthropic and OpenAI models are tested in parallel lanes with identical serves, rules, and questions. Each model’s single answer moves the ball one step, so the ball’s speed reflects the model’s response time without skipping or speeding up. The experiment shows that chat models perform poorly on this real-time task, highlighting the advantage of a non-waiting loop.