OpenClaw Model Shootout on Strix Halo (2026)

Six local AI models, five real-world homelab tests, one honest assessment. In Part 4 of the OpenClaw on Strix Halo series, I compare Qwen 3.5 (9B, 27B, 122B), GPT-OSS, GLM-4.7-flash, and Nemotron-Cascade-2 on actual monitoring tasks -- service crash diagnostics, morning briefs from RSS, script generation, capacity planning, and incident report writing. Full written guide: https://www.localhake.com/content/openclaw-local-model-shootout Recommended Gear Some links are affiliate links -- I may earn a commission at no extra cost to you. GMKtec EVO X2 (AI MAX+ 395, 128GB) - Featured in this video: https://amzn.to/4rJsga0 Minisforum MS-S1 MAX (AI MAX+ 395, 128GB): https://amzn.to/4uUZMg7 Beelink GTR9 Pro (AI MAX+ 395, 128GB): https://amzn.to/3PO5gt4 GEEKOM A9 Mega (AI MAX+ 395, 128GB): https://amzn.to/4lWatuZ Chapters 0:00 Intro 1:36 The Six Models 2:32 The Five Tests 3:31 Pulling Models & Switching in OpenClaw 3:57 Baseline Speed Benchmark 5:10 Test 1: Service Crash Diagnostics 5:49 Nemotron vs GLM: Pipeline Issues 7:04 Setting Up the Morning Brief Cron Job 9:45 Morning Brief Demo & Results 12:08 Test Results Overview 13:50 Honest Assessment & What's Next This is Part 4 of a 6-part series: Part 1: Agent Setup https://youtube.com/watch?v=_xKJE0Olid0 Part 2: AI Incident Response with Uptime Kuma https://youtube.com/watch?v=o9t7_1Mn8iE Part 3: Self-Hosted Matrix Command Center https://youtube.com/watch?v=MfD00eDPnxU Part 4: Local Model Shootout (this video) Part 5: Backup & Restore https://youtube.com/watch?v=2cpjPzEVscU Part 6: HTTPS & Clean URLs with Caddy https://youtube.com/watch?v=69UZ7HApkgI #homelab #ai #openclaw #strixhalo #localllm #selfhosted