Remix.run Logo
▲ verdverm 3 hours ago

that's a matter of hardware, Cerebras reports 1500 tok/s

https://inference-docs.cerebras.ai/models/qwen-3.8-27b

▲msikora 3 hours ago | parent [-]

Input $0.99 Output $1.49

That times 5 is steep!

▲verdverm 3 hours ago | parent [-]

Getting close to 80 tok/s on the dgx spark with NVFP4

https://spark-arena.com/leaderboard

(you'll have to filter model yourself, no query params)