Vera Rubin 3.8x output token throughput for Cognition RL workloads
Company: CoreWeave
The claim, verbatim
Cognition recorded a 3.8x boost in output token throughput for reinforcement learning workloads on Vera Rubin NVL72 on CoreWeave.
Source (primary)
CoreWeave Delivers NVIDIA Vera Rubin NVL72 at Scale - CoreWeave (-, news_article)
View cached copy (2026-10-05)Live source ↗Archive.org ↗
Quote: “the team recorded a 3.8 times boost in output token throughput for reinforcement learning workloads”
How we checked this
Checked on October 6, 2026. The cited source supports every part of this claim.
The release says that in the same benchmarks on CoreWeave Cloud, Cognition's team recorded a 3.8x boost in output token throughput for RL workloads on Vera Rubin NVL72.
Confirmed in the source:
- Cognition recorded a 3.8x boost in output token throughput
- The boost was for reinforcement learning workloads
- The result was on Vera Rubin NVL72 benchmarked on CoreWeave Cloud
What we did: Read our cached copy of the publisher (https://www.coreweave.com/news/coreweave-delivers-nvidia-vera-rubin-nvl72-performance-at-production-scale-starting-with-cognition) in full (13,996 characters, retrieved October 5, 2026) and checked each assertion in the claim against it.
Additional evidence
confirms CoreWeave Delivers NVIDIA Vera Rubin NVL72 at Scale - CoreWeave
Quote: “the team recorded a 3.8 times boost in output token throughput for reinforcement learning workloads”