I’m weighing a local DeepSeek setup against using a hosted model through an online service. Keeping sensitive prompts off a third-party server is appealing, but I’m unsure how practical local use is for someone without a high-end GPU.

If you’ve tried both, how noticeable is the trade-off in speed and answer quality? I’d also be interested in what hardware you consider a reasonable starting point.