CoreWeave Sets New Records in MLPerf Inference Benchmarks Using NVIDIA Blackwell Architecture

CoreWeave Inc. announced landmark results in the MLPerf Inference v6.0 benchmark suite on April 1, 2026. Utilizing NVIDIA’s newest GB200 and GB300 NVL72 infrastructure, the company secured top rankings in the Datacenter Closed division. These results highlight the platform's ability to translate raw compute power into superior real-world performance for the most advanced frontier AI models.

The benchmarks focused on highly demanding reasoning models, specifically DeepSeek-R1 and GPT-OSS-120B. CoreWeave led performance for DeepSeek-R1 in both server and offline modes, providing high throughput measured in tokens per second per GPU. The configuration showcased remarkable efficiency on the sparse Mixture-of-Experts architecture, which requires complex dynamic routing and high-bandwidth communication.

The transition to the NVIDIA GB300 NVL72 platform yielded dramatic efficiency gains for the company. CoreWeave recorded double the performance on the same hardware footprint compared to its own MLPerf 5.1 results from just six months prior. This rapid evolution underscores a commitment to full-stack optimization that bridges the gap between hardware capability and production output. Currently, eight of the ten leading AI model providers rely on the specialized cloud infrastructure provided by CoreWeave. This widespread adoption is supported by the company's recent achievement of a Platinum ranking in the SemiAnalysis ClusterMAX evaluations. Such independent validation confirms a lead in reliability and efficiency for large-scale distributed workloads.

This technical success follows a historic 8.5 billion dollar financing facility secured by the company in late March. The investment-grade rated debt marks a turning point for the industry, as traditional capital markets now view AI compute clusters as stabilized infrastructure assets. CoreWeave's public listing on the Nasdaq under the ticker CRWV further solidifies its institutional standing and growth potential.

As enterprises move AI from experimentation into production, inference performance has emerged as the defining metric of success. CoreWeave's latest benchmarks provide a blueprint for how purpose-built infrastructure can satisfy the rigorous demands of agentic workloads. The company remains positioned to scale its platform to meet the surging global demand for mission-critical AI services.