StepFun 3.7 Flash MTP Bench Strix Halo
Signal
75
Hype
15
In three linesStepFun Step-3.7-Flash benchmarked on AMD Ryzen AI Max+ 395 with MTP speculative decoding: +27.5% generation speed (26 tok/s vs 20.4), 84.7% draft token acceptance, -14% power consumption. No prefill degradation.Read source
Your take?
Summary generated by Claude — human-verified