Comment by guregodevo

Happy hallowen! ::::

Actually we tried Cerebras. It caps output at 32K tokens so the 128K reasoning (the author budget) didn't fit. Cerebras speed is real though