Show HN: Jev vs. GPT-5.6 and Claude Haiku at Pong
4 days ago
- Jev is a typed-decision model, not a chat model, designed for decision-making rather than conversation.
- The demo uses four lanes in Pong, each controlled by a different AI model, to compare their response times fairly.
- Each model's answer moves the ball one step, meaning slower models result in slower ball movement; no responses are skipped or sped up.
- Pong is highlighted as the Atari game that chat models handle worst, making it a fair challenge for the comparison.
- The four lanes feature well-known models, and the same question has also been tested on five additional models, including GPT-6 Astra and Claude Fable 5.1, with consistent results.
- The project is powered by TypeSafe AI and Vercel's AI Gateway, and it includes a recorded run with a replay feature.