AMD MI355X lands between Hopper and Blackwell in Artificial Analysis agent inference benchmark
ArtificialAnlys · x · 2026-10-15
Artificial Analysis launched AA-AgentPerf, a hardware benchmark for the agent era that replays real coding-agent trajectories, testing long context, tool-calling delays, and KV cache reuse; submissions are open with rolling results.
Key numbers on GLM-5.2/5.3:
- MI355X serves 26M tokens/s per MW at 100 tok/s per user — ahead of H200 (16M) but behind B200 (42M) and B300 (44M)
- Single-agent speed: MI355X matches H200 at 228 tok/s vs 312 on B200 and 333 on B300
- Coverage includes DeepSeek V4.1 Flash, DeepSeek V4 Pro, Kimi K3, and GLM-5.2/5.3 configurations
Related event: AMD MI355X Benchmarked on AA-AgentPerf: Mixed Efficiency Results(2 posts)→
More from Infra
- OpenAI infra lead leaves to start up after locking in double-digit GWs of compute — DesireeCachette · 2026-10-15
- Who Sells Data directory maps 159 AI training data vendors, compares Scale vs Mercor — cneuralnetwork · 2026-10-15
- Compute is the next commodity: building financial infrastructure for GPU markets — AnneliesGamble · 2026-10-15
- OCP slide reveals NVIDIA Rubin Ultra NVL576 will use CPO for scale-up networking — BenBajarin · 2026-10-15
- TensorFold ships macOS app for local LLM inference, reports 874 tok/s on RTX PRO 6000 — HankYeomans · 2026-10-15
- Enroute open-sources Git server hitting 7,600 concurrent pushes/s per repo on Raft and S3 — DanielLockyer · 2026-10-15