CoreWeave announced Forge and early-access NVIDIA Vera Rubin NVL72 availability on September 30. NVIDIA's corresponding account names Cognition, the company behind Devin, as the first customer running production workloads on Vera Rubin.
Connecting production with model development
Forge brings together Weights & Biases, OpenPipe post-training expertise and marimo notebooks. NVIDIA's account says ARIA and Sandboxes are generally available, while RL Rollouts remains in private preview. Availability therefore differs across components.
A customer benchmark with a specific baseline
Cognition reported up to 4.8 times the total token throughput for SWE-2 inference relative to GB200 NVL72, using sampled software-engineering tasks. This is a customer measurement on partner infrastructure, not an independent cross-cloud result. Capacity, regional access and performance under other workloads remain uncertain.