Comment by aeonfox

Separate to the self-host/datacentre argument, it would be interesting to see a speed/performance/watts-per-token leaderboard between leading models. Which model is the most watt-efficient?