hardicecream
Re:Show HN: A Codenames-style wordgame that benchmarks LLMs
on Tuesday November 17, 2026 @08:48PM (#50540858)
Sup guys,
I made a Codenames-style word game where you can play with and against LLMs. Feel free to check it out and play for free if you want against GPT Luna. In case you guys are curious, I have Fable 5.2 as the strongest model, but it seems to perform worse against non-Anthropic models. I guess that different companies have different training methods, so the results between cross-family matchups can be less predictable.
Let me know what you guys think if you have any feedback or confusion about the site, and I'll answer you right away.