AA-AgentPerf adds market GPU rental pricing to measure agent throughput per dollar
ArtificialAnlys · x · 2026-11-02
Artificial Analysis is partnering with Silicon Data to bring market-based GPU rental price indices into AA-AgentPerf, its datacenter system benchmark, reporting daily-updated value-for-money metrics: how much agent work each dollar of compute supports.
AA-AgentPerf tests inference hardware by replaying real coding agent trajectories — multi-turn sessions growing beyond 100K tokens with KV cache reuse and tool-calling delays — and compares systems like Vera Rubin, GB300/B300, GB200/B200, H200/H100 and MI355X on throughput per MW, per accelerator and per $/hr. The benchmark is open for submissions with rolling results.
More from Infra
- "My AI Runs Local": the sticker every AI laptop needs — DavidLinthicum · 2026-11-02
- US-China truce paperwork lags reality: AI supply chain deadlines hit from Nov 9 — demian_ai · 2026-11-02
- swift-transformers 1.4.0 ships with breaking decoding changes from 15 contributors — pcuenq · 2026-11-02
- France leads CVPR 2026 submissions in compute hours, credited to Jean Zay supercomputer — IgorCarron · 2026-11-02
- Podcast: AI verification hardware could enable a binding US-China AI treaty — wfithian · 2026-11-02
- ISTA releases 763B DeepSeek coder at 1 bit per param, pruning half the experts with 95% performance kept — Loginhe · 2026-11-02