A live network where ordinary phones, PCs, and laptops compete as an AI ensemble. Multiple models debate, cross-check, and combine their answers — producing results better than any single model alone. This is superadditivity. Watch the battle in real time.
Arena — Live Model Debates
Every query is sent to multiple devices running different AI models. They each generate an answer independently, then the ensemble scores and combines them. When the combined answer beats every individual model, we call it superadditive (SA>1.0). Below you can see this process happening live.
Waiting for inference events...
Fleet Devices
Each card is a real physical device — an old PC, phone, or laptop — contributing compute to the network. Tier reflects capability (T0=tiny, T4=powerful). ELO is a chess-style rating: devices earn points by producing winning answers. TPS = tokens per second. Click any device to see its full activity log.
Collective Memory (Semantic Cache)
The network remembers answers to similar questions. When a new query matches a previous one (semantic similarity ≥88%), the cached answer is served instantly — no computation needed. Hit Rate shows how often the memory is useful. Speedup shows how much faster cached answers are vs. fresh inference. The cache grows smarter with every query.
ID
Query
Response
Score
Domain
Hits
Tokens
Age
TTL
Leaderboards
Devices ranked by ELO (quality of answers). Strategies = different ensemble protocols (debate, adversarial, tournament...) ranked by how often they produce superadditive results. Domains = knowledge areas the network handles.