Comment by perfmode

How’s the token throughput / response time?