# Artificial Analysis releases open-source tool to benchmark local AI agents on laptops and workstations

Digest AI · Hardware & Compute · published 2026-09-30T08:41:50Z

Canonical: https://digestai.news/story/artificial-analysis-releases-open-source-tool-to-benchmark-local-ai-ag

## Summary

Artificial Analysis has launched **AA-AgentPerf-Local**, an open-source benchmarking tool designed to measure how fast AI agents perform on laptops and workstations. The tool replays real agent trajectories—eight recorded tasks spanning 168 model turns—with a context window growing to ~56K tokens. It isolates inference speed by default but can also simulate tool delays or live CPU calls.

The initial results cover four hardware setups: **NVIDIA DGX Spark (128GB)**, **RTX 5090 (32GB)**, **AMD Ryzen AI Halo (128GB)**, and **MacBook Pro M5 Pro (64GB)**. Benchmarks focus on models like **Qwen3.5-9B, Qwen3.8-27B, Qwen3.6-35B-A3B, and Ling 3.0 Flash (124B/5B active)**, all tested at 4-bit quantization. The **RTX 5090** emerged as the fastest system for models fitting in its memory, while the **DGX Spark** outperformed the Ryzen AI Halo due to higher compute efficiency. The MacBook Pro (M5 Pro) showed competitive results, finishing within 2–21% of the Ryzen AI Halo for some models. The tool also highlights how prefill speeds—reading new input tokens—can dominate latency, especially on systems with lower compute-to-memory bandwidth ratios. Future updates will expand hardware coverage, add multi-agent scenarios, and include user-submitted leaderboards.

## Key points

- AA-AgentPerf-Local benchmarks local AI agent performance on laptops/workstations with 8 recorded tasks (~56K tokens)
- RTX 5090 fastest for models fitting in 32GB; DGX Spark outperforms Ryzen AI Halo due to compute efficiency
- Tool supports any OpenAI-compatible server and will expand to multi-agent scenarios and live CPU tool-calling

## Why it matters

Helps developers and businesses compare hardware and model setups for local AI deployment, reducing trial-and-error costs. Useful for small teams running agents on laptops or workstations without cloud dependencies.

## Sources

1. [AA-AgentPerf-Local: Benchmarking local AI agents on laptops and workstations](https://artificialanalysis.ai/articles/aa-agentperf-local) (artificialanalysis.ai, 2026-09-30)

Part of the developing story: [Artificial Analysis AI Benchmarking Initiative Chronicle](https://digestai.news/thread/artificial-analysis-launches-cyber-index-alliance-to-benchmark-ai-agents-on) (2 stories)

## Cite

Digest AI, "Artificial Analysis releases open-source tool to benchmark local AI agents on laptops and workstations", 30 September 2026, https://digestai.news/story/artificial-analysis-releases-open-source-tool-to-benchmark-local-ai-ag

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/artificial-analysis-releases-open-source-tool-to-benchmark-local-ai-ag.json