Jev vs. GPT-5.6 and Claude Haiku at Pong
Jev returns a decision in 227 ms. The chat models take 2.5 to 3.5 seconds. Pong where the ball moves one step per model decision. Slow model, slow ball.
Jev vs. GPT-5.6 and Claude Haiku at Pong launched on September 18, 2026 with 10 votes, #925 of 8,042 launches that month and more than 90% of that year's launches.
In plain words
Jev vs. GPT-5.6 and Claude Haiku at Pong is a comparison tool that plays the classic Atari game simultaneously with different AI models. Each model controls a paddle in its own lane, and the ball moves one step per model response, making response speed directly visible. Jev, a typed-decision model, completes decisions in 227 milliseconds, while chat-based models take 2.5 to 3.5 seconds, resulting in noticeably slower gameplay. The tool is for developers and researchers examining real-world performance differences between model architectures.
written from the facts on this page · September 2026
From the sources
Four lanes, one game: same serve, same rules, same question, asked of a different model in each. One answer moves the ball one step, so every ball travels at its own model's answer time and nothing is skipped or sped up. Jev is a typed-decision model rather than a chat model, and Pong is the Atari game chat models handle worst , so a loop that won't wait is the fair place to see what that'sfrom jev-pong.ably.dev
Does a similar job
all alternatives →
I taught AI to commentate Pong in real time2025 · github.com · ▲207LLM-generated real-time commentary for Pong. Contribute to pncnmnp/xpong development by creating an account on GitHub.
Getting GLM 5.2 running on my slow computerJul 2026 · github.com · ▲937A few days ago I found myself trying out GLM 5.2 and was really positively impressed. The capabilities and security I was getting from this LLM are similar to those I've gotten from models like Claude or GPT, and this really surprised me. But then I thought, "I wonder how it would work on a normal computer like mine," and above all, "I wonder if it would work without going into OOM on a computer like mine." So I started working with the help of agents to test this possibility. I started converting the model to int4, understanding MTP usage, and if possible implementing DSA for long context.…
Real-time AI Voice Chat at ~500ms Latency2025 · github.com · ▲524Have a natural, spoken conversation with AI! Contribute to KoljaB/RealtimeVoiceChat development by creating an account on GitHub.
GPT‑5.4 mini and nanoMar 2026 · openai.com · ▲267Fast and efficient models optimized for coding and subagents
Jevstiller – Distill Jev into a local model, with a disagreement bound3d ago · jevstiller.pages.dev · ▲66September 2026. Every number here is from the benchmarks, and bash experiments/bench.sh --no-record reruns them without an API key.
More ai this month
the category →
E-ink bird frame for Raspberry Pi - real-time bird detection by audio, fully local AI, rendered as real, hand-cut 1800s bird illustrations. - arnegiacomo/fugleramme
AI · 17d ago · github.com
Launched alongside, September 2026
the whole month →E-ink bird frame for Raspberry Pi - real-time bird detection by audio, fully local AI, rendered as real, hand-cut 1800s bird illustrations. - arnegiacomo/fugleramme
AI · 17d ago · github.com
- WI
Special relativity simulator in which light crawls along at a human speed, so Lorentz contraction, aberration and Doppler shift happen at walking pace.
Life & fun · 23d ago · rivendell.dmitrybrant.com
Start coding agents, control sessions, review PRs. Anywhere.
Dev tools · 22d ago · kilo.ai