DeepSeek on 4 DGX Sparks codes great with Opus 5.5 planning, GLM-5.3 not ready
AIFlow_ML · x · 2026-11-23
@sonderby ran DeepSeek V4.1 Flash on 4× NVIDIA DGX Spark as an OpenCode agent, building two creative web apps three ways:
- DeepSeek alone
- Opus 5.5 plans → DeepSeek codes
- GLM-5.3 fully local (2.8-bit, same Sparks) plans → DeepSeek codes
Verdict: a good plan makes the local coder faster and better, but a flawed plan gets coded exactly as written. In his small experiment GLM-5.3 wasn't ready as a planner at Opus 5.5's level, so 'local coder, cloud planner' remains the practical line — and the open question is which local model can fill the planner role.
More from coding & agent
- Popfold: an open-source three.js engine built with parallel Claude Code sessions — cherichita · 2026-11-23
- Weaviate Podcast #150 dives deep into rerankers and agentic search — CShorten30 · 2026-11-23
- Job-hunting agent fixes reviewed: idempotency keys help, but AUC stays 0.645 — nivokvo · 2026-11-23
- Inference practitioner on GPU co-ops: saturation math, weight caching, and quantization as dark art — sloppenheimer · 2026-11-23
- Siemens and AGIBOT generate robot grippers from text via agentic AI workflow — emmanuelvivier · 2026-11-23
- Made a Video for $0.09: Full Workflow with Claude + VidLab API — eptwts · 2026-11-23