19 LLMs Battle Through 376 Rounds of Codenames-Style Word Game

A developer ran 376 rounds of a Codenames-style word game with 19 LLMs paired into teams. Gemini 3.8 Flash topped guessing accuracy at 93%, and the experiment found guessing skill matters more than clue-giving.

2026-10-31 ~ 2026-10-31 · 2 related posts