AA-AgentPerf-Local launches as an open-source benchmark for local AI agents

Artificial Analysis says AA-AgentPerf-Local measures how quickly local systems run recorded agent tasks, skipping tool execution by default to isolate inference speed. In its initial tests, the GeForce RTX 5090 was more than 3.5 times faster than the other systems on every model that fit in its 32 GB. The group plans to expand testing to more hardware, models and multi-agent setups.