Cognition runs SWE-2 on NVIDIA Vera Rubin, citing 4.8x per-GPU throughput over GB200

Cognition is the first customer running NVIDIA Vera Rubin in production, hosted by CoreWeave, according to SGLang's October 9, 2026 post on X. SGLang says Cognition has served its SWE-2 model on Rubin with SGLang since September.

SGLang says Rubin delivers about 4.8x more token throughput per GPU than GB200 on SWE-2 inference, at the same decode speed. The figure comes from SGLang's post and has not been independently verified.