Continue, M4 Pro Fuel Local LLM Debate

For most developers, the cloud subscription is still the better default; local wins when privacy, offline access, or tinkering matter more than raw model quality.

• –Continue officially supports local and offline configurations, so the workflow is real rather than a science project.
• –Apple's M4 Pro gives you 48GB unified memory and 273GB/s bandwidth, which makes mid-size quantized coding models practical.
• –Commenters think Qwen-class models are surprisingly usable for autocomplete, lightweight refactors, and private side projects.
• –Cloud models still dominate harder agentic tasks, longer context, and cases where latency or reliability matters more than privacy.
• –The best fit is often hybrid: local for sensitive code and quick suggestions, cloud for the hard thinking.