AA-AgentPerf-Local hits GitHub: benchmark your machine's agent inference speed

ArtificialAnlys · x · 2026-09-30

Artificial Analysis published the GitHub repo and initial results for AA-AgentPerf-Local. The open-source tool replays recorded agent conversations to measure local LLM serving throughput and latency, works with any OpenAI-compatible server or auto-downloads a pinned model, and supports macOS, Linux, and Windows (managed runs need NVIDIA GPU).

Original post →

More from Infra

Infra channel →