"Zambezi team on AMD Developer Hackathon: ACT II Hackathon"

Team Idea

Blind A/B arena for LLMs. Prompt two anonymous models head-to-head, best-of-5, full reveal of identity + cost + tokens at the end.

Team Members

AMD ZAMBEZI

fx4100

Submission

Blind Model Arena

Blind Model Arena is a side-by-side LLM benchmarking platform. Compare responses blindly to eliminate brand bias, track real-time token/TPS performance, and play 'Catch the Speeder': a speed-guessing minigame comparing cloud vs local AMD GPU(s).