Artificial Analysis open-sources local agent inference benchmark
What happened
On September 29, 2026, Artificial Analysis open-sourced AA-AgentPerf-Local, a tool that replays real agent traces to test local inference performance on laptops and workstations. It published preliminary results for NVIDIA DGX Spark, GeForce RTX 5090, AMD Ryzen AI Halo and MacBook Pro M5 Pro. On September 30, coverage added that the tool replays 8 real agent tasks, 168 rounds of model interaction in total, with context growing to about 56K tokens. So far the story is the release and open-sourcing of the tool plus those preliminary results.
Written by AI from the coverage · updated 47 minutes ago
Coverage
Follow the reports to see the story from different sides.
- AI HOT · ProductsArtificial Analysis 开源 AA-AgentPerf-Local 本地智能体推理测试工具
Artificial Analysis 发布开源工具 AA-AgentPerf-Local，用于测试本地推理性能。该工具在笔记本和工作站硬件上回放 8 个真实智能体任务、共 168 轮模型交互，上下文增长至约 56K tokens。
- AI HOT · ProductsArtificial Analysis 开源 AA-AgentPerf-Local，测试笔记本与工作站本地 AI 智能体推理性能
Artificial Analysis 开源 AA-AgentPerf-Local，通过回放真实智能体轨迹测试笔记本与工作站上的本地推理性能，并公布 NVIDIA DGX Spark、GeForce RTX 5090、AMD Ryzen AI Halo 和 MacBook Pro M5 Pro 的初步结果。
Heat over time
Not enough continuous observations to draw a trend yet.