AsariAI's Self-Improving Agents Boost vLLM Throughput by 16% on B200s

yisongyue · x · 2026-07-30

Startup AsariAI showcased a breakthrough using their self-improving agents to optimize the full vLLM inference stack.

Running on B200s, the agents achieved up to 16% more throughput and interactivity for DeepSeek v4 Pro and Zai.org's GLM 5.2 (without MTP). The company noted that every code change was verified, and the agents became progressively better and faster at optimizing with each iteration.

Related event: AsariAI's Self-Improving Agents Boost vLLM Throughput by 16%(3 posts)→

Original post →

More from coding & agent

coding & agent channel →