NVIDIA Hermes Agent Hits 140K GitHub Stars: Self-Evolving Skills on DGX Spark with Qwen 3.6 at 1 PFLOP

NVIDIA Blog·medium signal

NVIDIA published a developer blog on May 13 detailing the Hermes Agent framework by Nous Research, which crossed 140,000 GitHub stars in under three months and ranks as the most-used agent globally on OpenRouter. Hermes writes and refines its own skills from experience, persists memory across sessions, delegates to short-lived sub-agents for parallel sub-tasks, and runs scheduled workflows via built-in cron. On NVIDIA DGX Spark (128GB unified memory, 1 PFLOP AI performance), Hermes runs Qwen 3.6 35B on ~20GB while surpassing 120B-parameter models — the 27B variant matches accuracy of 400B models at 1/16th the size.