CoreWeave Leads MLPerf 0.7 Endpoints Benchmark With DeepSeek-R1
CoreWeave, Thursday, September 10th, 2026
Leading per-GPU DeepSeek-R1 throughput among NVIDIA GB200 NVL72 submissions, tested on production infrastructure.
CoreWeave reports the leading per-GPU DeepSeek-R1 throughput among NVIDIA GB200 NVL72 submissions in the inaugural MLPerf 0.7 Endpoints benchmark.
The detail worth noting is that the results were produced on production infrastructure rather than a purpose-built benchmark configuration, which makes the numbers more representative of what a customer would see.
The new Endpoints category itself is useful, since it measures serving performance under realistic request patterns rather than raw throughput on a fixed batch.