GB10 agent bench
Jun 20, 2026
·
1 min read
Benchmarking a fully local agent stack (Ollama models + OpenHands) against Claude Code and Codex CLI on the same tasks, on an ASUS GB10 Grace Blackwell box. Tracking where local models hold up and where they don’t.
Authors
Digital IC Design Engineer · Homelab Operator
Digital IC Design engineer with 20+ years of experience — design, verification, scripting, automation, LEC, LINT, SYN. Open to work in Japan/Singapore.
Off the clock I run a home network that’s grown well past reasonable size: a GB10 AI box, a mesh of UniFi and AmpliFi gear, three cameras streaming live, and a small fleet of always-on services that occasionally page me at 2am.