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

Category: ai-ml

Tags: latency-benchmark, llm-agents, realtime, pong, ably, vercel-ai-sdk, multiplayer

Score: 6.5/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 4)

Jev Pong is a latency benchmark disguised as a Pong game: four lanes each run a different model as the right paddle, where one model decision moves the ball one segment. It showcases a typed-decision model (Jev) that answers in ~227ms versus 2.5-3.5s for chat models using structured output, and demonstrates a live multi-agent pattern where the agent is a presence member of the same Ably channel as the player rather than behind a request.

Target audience: AI engineers, realtime/backend developers, agent framework builders

Repository: https://jev-pong.ably.dev/ · TypeScript · Apache-2.0

View on Hacker News