SemiAnalysis · 算力产业· Bryan Shan·· 3 hours agoSelectedAI score77
SemiAnalysis Releases Initial TPUv7 Benchmark Results in AgentX Agentic Inference Evaluation
The Iron(wood) Throne: Google's TPU Stakes Its Claim on Inference
AI guide
In its InferenceX/AgentX agentic inference benchmark, SemiAnalysis published the first multi-turn, long-context inference evaluation data for Google TPUv7 (Ironwood), covering four models: Meta Muse Spark Open, GLM-5.2, Qwen (Qwen3.5 397B), and Kimi K3.
Why it matters
The benchmark showcases TPUv7's real-world cost and performance in long-context, multi-turn agentic inference, offering a practical reference for compute selection.
Source:SemiAnalysis · 算力产业 · newsletter.semianalysis.com