Vera Rubin inference throughput gain vs GB200 (Cognition benchmark)
Company: CoreWeave
The claim, verbatim
Cognition measured up to a 4.8X increase in total token throughput for SWE-2 inference on Vera Rubin NVL72 versus a GB200 NVL72 baseline, in what CoreWeave described as the first customer-executed inference benchmark on the platform.
Source (primary)
CoreWeave brings Nvidia Vera Rubin, AI tools to its cloud services - Network World (-, news_article)
View cached copy (2026-10-02)Live source ↗Archive.org ↗
Quote: “it measured up to a 4.8X increase in total token throughput for SWE-2 inference workloads on Vera Rubin NVL72 compared with an Nvidia GB200 NVL72 baseline”
How we checked this
Checked on October 3, 2026. The cited source supports every part of this claim.
The article gives the 4.8X figure, the GB200 NVL72 baseline, and CoreWeave's description of it as the first customer-executed inference benchmark.
Confirmed in the source:
- Cognition measured up to a 4.8X increase in total token throughput for SWE-2 inference workloads
- The comparison was Vera Rubin NVL72 against an Nvidia GB200 NVL72 baseline
- CoreWeave described this as the first customer-executed inference benchmark on the platform
What we did: Read our cached copy of the publisher (https://www.networkworld.com/article/4229739/coreweave-brings-nvidia-vera-rubin-ai-tools-to-its-cloud-services.html) in full (8,444 characters, retrieved October 2, 2026) and checked each assertion in the claim against it.
Additional evidence
confirms CoreWeave brings Nvidia Vera Rubin, AI tools to its cloud services - Network World
Quote: “it measured up to a 4.8X increase in total token throughput for SWE-2 inference workloads on Vera Rubin NVL72 compared with an Nvidia GB200 NVL72 baseline”