AA-AgentPerf launches live results page for agentic inference benchmark across B200 to MI355X
ArtificialAnlys · x · 2026-10-14
Artificial Analysis added a live results page for its AA-AgentPerf agentic inference benchmark, detailing the methodology: each simulated agent replays a real multi-turn coding session with context compounding past 100K tokens, prior turns served from KV cache and only new tokens prefilled. Results are viewable by throughput per MW, per $/hr, and per accelerator across DeepSeek V4.1 Flash, V4 Pro 0813, Kimi K3, and GLM-5.2/5.3 on hardware from B200 to MI355X, with open submissions.
More from Infra
- Agent traffic is a strange inference workload: long prefix caching beats raw tokens per second — rohanpaul_ai · 2026-10-14
- LightVAE speeds up video VAE decoding up to 2.24x on H100 with minimal quality loss — Shiqiao Gu · 2026-10-14
- Orbital data centers have no off-switch: Reddit debates AI safety paradox — Zappy_Oh · 2026-10-14
- Rednote's Dots unveils MSAR: 1.58x iso-loss compute speedup over residual baseline — teortaxesTex · 2026-10-14
- VirusTotal ships macOS Mirage sandbox to analyze fake password-prompt infostealers on Apple Silicon — jedisct1 · 2026-10-14
- ClickHouse merges IPCrypt support for GDPR-friendly IP pseudonymization — jedisct1 · 2026-10-14