Vera Rubin 4.8x token throughput vs GB200 for Cognition
Company: CoreWeave
The claim, verbatim
In Cognition's early tests, Vera Rubin NVL72 delivered up to a 4.8x increase in total token throughput for SWE-2 inference over GB200 NVL72.
Source (primary)
From Training to Production, NVIDIA and CoreWeave Close the Loop on Agentic AI - NVIDIA Blog (-, news_article)
View cached copy (2026-10-01)Live source ↗
Quote: “Cognition saw Vera Rubin NVL72 deliver up to a 4.8x increase in total token throughput for SWE-2 inference workloads over GB200 NVL72.”
How we checked this
Checked on October 12, 2026. The cited source supports every part of this claim.
The article reports Cognition's early tests showing up to 4.8x higher total token throughput on Vera Rubin NVL72 versus GB200 NVL72 for SWE-2 inference.
Confirmed in the source:
- The result came from Cognition's early tests
- Vera Rubin NVL72 delivered up to a 4.8x increase in total token throughput
- The comparison was for SWE-2 inference workloads against GB200 NVL72
What we did: Read our cached copy of the publisher (https://blogs.nvidia.com/blog/coreweave-agentic-ai-vera-rubin/) in full (9,968 characters, retrieved October 1, 2026) and checked each assertion in the claim against it.
Additional evidence
confirms From Training to Production, NVIDIA and CoreWeave Close the Loop on Agentic AI - NVIDIA Blog
Quote: “Cognition saw Vera Rubin NVL72 deliver up to a 4.8x increase in total token throughput for SWE-2 inference workloads over GB200 NVL72.”
confirms First Vera Rubin NVL72 Customer Sees 4.8x Throughput - CoreWeave
Quote: “our engineers are seeing up to a 4.8x increase in total token throughput for SWE-2 inference workloads over GB200 NVL72”
confirms CoreWeave Delivers NVIDIA Vera Rubin NVL72 at Scale - CoreWeave
Quote: “Cognition measured a 4.8 times increase in total token throughput for its SWE-2 inference workloads on NVIDIA Vera Rubin NVL72 compared to a GB200 NVL72 baseline”