Jevman launches AI Pac-Man decision benchmark
What happened
On October 9, a Show HN post from Jevman appeared on Hacker News' front page. The open-source benchmark pits AI decision models against arcade ghosts in classic Pac-Man, running 100 games per model and ranking by average score, with scores within a 95% error range treated as tied. Hosted, fine-tuned or local models can connect through an HTTP endpoint. At each intersection the game sends maze state as JSON and requests a direction; the model returns probabilities per direction, and if it times out after 2 seconds a fallback rule takes over. Submitted results are replayed and verified in CI, then added to the leaderboard as self-reported scores.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageShow HN: Jevman – AI decision models play Pac-Man
Jevman 让 AI 决策模型在经典吃豆人游戏中与街机幽灵对战，每个模型各跑 100 局并按平均分排名，分数在 95% 误差范围内视为并列。该基准开源，任何部署在 HTTP 端点后的模型（托管模型、微调模型或本地模型）都可接入：游戏在每个路口以 JSON 发送迷宫状态并请求方向，模型返回各方向概率，超时 2 秒则由备用规则接管。提交结果经 CI 回放校验后以自报成绩加入排行榜。
Heat over time
Heat now 7·Comparable peak 8（Oct 9 09:00）·Comparable change over 24 hours –
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.