Jevman: AI Decision Models Play Pac-Man
opper.ai- Category
- AI Tools
- Rank
- No. 1593Tools index
Previous survey · No. 1558 ·
- Type
- GAME
- Use case
- Model & Agent Evaluation
- Interfaces
- Web
- GitHub
- 22 stars
- Date
About
Opper ran several low-latency AI decision models (including jev 1.13, kev, clef, clef flash, and GPT-6 Luna and Laya) through 100 Pac-Man games each against bot ghosts, using Pac-Man as a benchmark for fast, simple decision making. Results are published on a leaderboard, and the repo is open-sourced so others can submit their own models. Visitors can also play as Pac-Man against ghosts controlled by the models, at about 2 cents per game, with free credits offered.
At a glance
JevMan uses Pac-Man to compare AI models that return probabilities over possible moves. It is worth opening if you want a concrete example of how a decision model behaves when it must choose an action before a deadline.
Good fit
Readers comparing specialized decision models and curious about how latency affects an interactive task.
Consider something else when
Skip it if you need a benchmark for coding, chat quality or decisions outside arcade gameplay.
Start here
Open the leaderboard and compare two ranked models on score and latency. Check who ran each entry and look for a tie marker before treating a small score gap as meaningful.
Cost and requirements
The leaderboard is public. It reports 100 games per model against the classic ghosts and separates ranked decision models from chat and rules-based references.
Limits to know
The benchmark measures one game. The publisher says it replays community submissions to check scores, while their latency and cost figures are supplied by submitters.
Sources checked 2026-11-05. Source-based assessment.
Tags
Tech Stack
Media

Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.