PALO ALTO — Inflection AI, in collaboration with CoreWeave and NVIDIA, has achieved a major milestone in AI computing, setting a new industry benchmark on MLPerf, the leading open-source standard for AI performance evaluation. Utilizing over 3,500 NVIDIA H100 Tensor Core GPUs, the Inflection AI cluster completed the reference training task in a record-breaking 11 minutes, the fastest performance for training large language models (LLMs) on this benchmark to date.

The achievement underscores the capabilities of Inflection AI’s state-of-the-art infrastructure, which is already among the largest computing clusters globally. The company plans to expand its size significantly in the coming months to further advance its AI development capabilities.

In a joint submission to MLPerf with CoreWeave and NVIDIA, Inflection AI demonstrated optimized performance through close collaboration and fine-tuning of its computing cluster. MLPerf, the industry-standard benchmark, evaluates AI workloads for training and inference, providing valuable insights into cutting-edge model performance.

This success follows the recent unveiling of Inflection-1, Inflection AI’s proprietary large language model. Inflection-1 has outperformed prominent models in its compute class, such as GPT-3.5, LLaMA, Chinchilla, and PaLM-540B, across a range of industry benchmarks. The model powers Pi, Inflection’s personal AI, enabling users to engage in seamless, natural conversations and access relevant, reliable, and fast information.

Inflection AI’s advancements are part of its broader mission to create safe, powerful, and accessible AI products for millions of users. With further milestones on the horizon, the company remains dedicated to setting new standards in AI development and user experience.