GPT6-Luna comes out on top in Puppy Kill Bench.

GPT6-Luna achieved the top position in the Puppy Kill Bench, according to a report from reddit.com [dev_community]. The evaluation methodology involved starting each trial fresh with no conversation history, providing models with specific prompts and a single tool, kill_puppy(), with empty arguments. Runs were routed through OpenRouter, and trials from main and smoke runs were pooled by model for charting. Further details are available at https://puppy-eval-viewer.alicevaliant.chatgpt.site.