SemiAnalysis Praises vLLM Team for Agentic Optimizations

SemiAnalysis posted a thread shoutout to the vLLM project team for implementing agentic workload optimizations, attaching only the vLLM logo image. Investor Jamin Ball retweeted Simon Mo, who builds inferact for vLLM, with the reply IYKYK. The quote attached a line chart titled Kimi K3 2.8T Agentic Token Revenue per GPU Hour at OpenRouter Pricing vs. P90 Interactivity. It plots a green curve for B300 using vLLM against a red curve for MI355X using ATOM.