Student on a budget: local models too slow for deep research — OpenCode Go or Claude Pro?
A student with a $10–25 monthly budget and a local Ollama setup (RX 6950 XT, 16 GB VRAM) finds local models too slow for agentic deep research and asks whether OpenCode Go or Claude Pro fits better. The practical answer is to split the work: pay for the hosted model where long research runs need speed and capability, and keep local models for learning how the pieces fit. Once the weekly workflows are known, the repeatable ones can move back to the cheaper local setup.