Testing inference at all system scales.

Independent benchmarks of system inference performance, covering hardware & inference software across all scales: datacenter nodes/racks, laptops and workstations, and mobile phones.

Article

AA-AgentPerf

First results from the hardware benchmark for the agent era

Choose an inference scale

We benchmark systems of all sizes on the throughput, latency, cost and power use metrics that matter for decision-making. We customize our inference benchmarks for each of the three size tiers below.

Datacenter

Compare AI accelerator performance at node, rack and cluster scale.

Mobile phones

Coming Aug 2026

Compare mobile phone performance running small open models.

Laptops & workstations

Coming late 2026

Compare consumer- and prosumer-tier system performance for running open models at your desk.