MLPerf Inference v6.1 Press Briefing Q3 2026

On September 10, 2026, MLCommons held the press briefing for MLPerf Inference v6.1. David Kanter (MLCommons founder and Head of MLPerf), Miro Hodak (Inference WG co-chair), Ramesh Chukka (End-to-End RAG task force chair), and Palanivel Guruva Reddiar (Agentic Edge Inference task force chair) presented the latest results and introduced two new benchmarks. Highlights: 486 datacenter and edge results from 30 organizations gpt-oss-120b is the most popular workload in MLPerf history First End-to-End RAG benchmark - real-world retrieval pipelines First Agentic Edge Inference benchmark - on-device agentic AI 2.7X-5.7X per-accelerator gains in one year Cisco's first cross-vendor submission (NVIDIA + AMD) Crusoe's 512-accelerator run, the largest ever NVIDIA Vera Rubin NVL72 preview debut Results: https://mlcommons.org/visualizer Datacenter: https://mlcommons.org/benchmarks/infe... Edge: https://mlcommons.org/benchmarks/infe... #MLPerf #MLCommons #AI #MachineLearning #Benchmarks #Inference #LLM #RAG #AgenticAI