Arm and Qualcomm push their own AI agent benchmarks

Arm published a CPU performance evaluation methodology for agentic AI on October 13, according to digitaltoday.co.kr, as chip companies move to set their own standards for measuring how well hardware handles AI agents.

Arm said its methodology rests on three principles: real agent use cases, repeatable measurement, and verification of the final task results. Alongside the methodology, the company released performance verification results from the US testing organization Signal65, which compared the Arm AGI CPU with AMD server chips. According to Arm, the AGI CPU delivered 1.99x the average performance per rack of the AMD EPYC 9755, and the time to open an isolated sandbox for an agent was up to 45% shorter than with the AMD EPYC 9965.

Qualcomm puts a new benchmark on the floor

Qualcomm introduced the Evolve GPU/AI benchmark from Dutch developer Traverse Research and ran measurements at the Snapdragon Summit 2026 benchmarking session, where the Snapdragon 8 Elite Extreme 6th Gen was among the chips tested. Traverse Research said Qualcomm and Samsung are currently running Evolve internally.

Traverse Research stated that no company owns or operates it and that performance cannot be bought with money.

Benchmark figures have drawn scrutiny before: at the Snapdragon Summit 2026 benchmarking session, measured AI benchmark results for the Snapdragon 8 Elite Extreme 6th Gen differed from Qualcomm's expectations depending on the measurement tool. Qualcomm has said AI performance varies with end-to-end optimization, software and the type of model running, so benchmarks alone cannot fully verify platform improvements. Intel has also commissioned a Signal65 report on how AI agents execute instructions, edit files and compile code between model answers.