[wyre]

It depends. If I was running a model locally I would much prefer 9000 t/s. If I was running an inference company, obv 1000 concurrent requests at 150 t/s is preferable.