AI decision models play Pac-Man: jev 1.13 tops leaderboard

A Reddit user benchmarked six decision models by having them play Pac-Man in real time and published an open-source leaderboard. jev 1.13 led with an average score of 2,750, followed by GPT-6 Luna and Clef Flash. Users can also play as Pac-Man against model-controlled ghosts.

• Cloudflare's Clef model removes humans from AI agent loop2026-10-02
• OpenAI Decisions API Confidence Unreliable2026-10-02
• Cloudflare open-sources Clef decision models, launches RL fine-tuning2026-10-01
• Four open decision models benchmarked on an RTX 40902026-10-08
• Clef Flash plays Snake in real time on RTX 50802026-10-05