Results
Menu

Benchmarks

Every MLCommons benchmark, its newest results, and where to explore them.

Latest benchmarks

MLPerf Endpoints

Endpoints v1.0

Released Nov 13, 2026 · 23 results

One submission: throughput vs. interactivity

NVIDIA · NVIDIA GB300 NVL72 x8 (TRT-LLM + Dynamo) · GPT-OSS 120B

Throughput (tok/s) ↑

050,000100,000150,000200,0000200400600Interactivity (tok/s/user) →164,189 tok/s at 13.4 tok/s/user153,554 tok/s at 25 tok/s/user131,353 tok/s at 42.8 tok/s/user93,697 tok/s at 91.5 tok/s/user80,035 tok/s at 156.3 tok/s/user57,091 tok/s at 223 tok/s/user31,412 tok/s at 245.4 tok/s/user18,940 tok/s at 295.9 tok/s/user15,810 tok/s at 329.4 tok/s/user12,157 tok/s at 379.9 tok/s/user7,559 tok/s at 472.5 tok/s/user
Open results Results table

All benchmarks

13 benchmarks · dates and counts only for results this site loads
MLCommons programBenchmark# of resultsResults tableExplore
Nov 13, 2026MLPerf EndpointsEndpoints v1.0 Latest23View tableOpen Visualizer
Jun 26, 2026MLPerf EndpointsEndpoints v0.7 10View tableOpen Visualizer
—AILuminateJailbreak —Not Available YetNot Available Yet
—AILuminateSafety (English) —Not Available YetNot Available Yet
—MLPerf InferenceInference: Datacenter —Not Available YetNot Available Yet
—MLPerf InferenceInference: Edge —Not Available YetNot Available Yet
—MLPerf ClientClient —Not Available YetNot Available Yet
—MLPerf MobileMobile —Not Available YetNot Available Yet
—MLPerf TrainingTraining HPC —Not Available YetNot Available Yet
—MLPerf StorageStorage —Not Available YetNot Available Yet
—MLPerf AutomotiveAutomotive —Not Available YetNot Available Yet
—MLPerf TinyTiny —Not Available YetNot Available Yet
—AlgoPerfTraining Algorithms —Not Available YetNot Available Yet