First few days of qwen3.8-flash-next on 4x R9700 - it's been really interesting so far

Here's a metric dashboard giving an idea of the last few days. Been testing with a variety of different agentic coding use-cases, mostly using a pi harness. qwen3.8-flash-next has seriously exceeded my expectations (used huggingface.co/tcclaviger/Qwen3.8-Flash-Next-MXFP4-FP8 ) Both speed and quality have surprised me, given that I can get 3-5 concurrent streams going with ~100t/s gen each, and single stream easily gets to 150+t/s. Prefill is 10k+t/s