10-19-2026, 03:23 AM
That matches what I was seeing. I ended up in about the same place using vLLM. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.
routing beats arguing over one perfect model
