CoreWeave Inc. (NASDAQ: CRWV) announced results from the MLPerf Inference v6.0 benchmark suite, participating in the Datacenter Closed division using NVIDIA GB200 NVL72 and NVIDIA GB300 NVL72 systems.

The company submitted benchmark results for two AI models: DeepSeek-R1 and GPT-OSS-120B. CoreWeave led performance for DeepSeek-R1 in server and offline modes when measured in tokens per second per GPU, according to the company's press release.

Using the NVIDIA GB300 NVL72 configuration, CoreWeave reported achieving 2X its own MLPerf 5.1 results on the same hardware footprint for DeepSeek-R1 server performance. The benchmark tested the systems' ability to handle DeepSeek-R1's sparse Mixture-of-Experts architecture.

"Inference is the defining layer in AI. It's where models are actually put to work and where performance in production shows up," said Peter Salanki, co-founder and chief technology officer of CoreWeave.

The MLPerf benchmark suite measures AI system performance across standardized tasks. CoreWeave's submissions used NVIDIA's reference configurations as the baseline for testing.

CoreWeave previously received Platinum rankings in both SemiAnalysis ClusterMAX 1.0 and 2.0 evaluations, which assess AI cloud performance and reliability metrics.

The company, established in 2017, completed its public listing on Nasdaq in March 2025. CoreWeave provides cloud infrastructure services focused on AI workloads and operates data centers equipped with specialized hardware for machine learning applications.