deepseek v4 pro on opencode

deepseek v4 pro on opencode is smart, but it's not intuitive for planning.

if you're coming from opus or sonnet, you'll notice the difference immediately. it follows instructions well and executes cleanly. it just doesn't plan or brainstorm with you the way the bigger reasoning models do.

i've tried gemma4 and qwen locally too. it's okay, but the m3 ultra spins up and heats up fast. opencode flash v4 is $10/month, and i use it for hermes and the homelab. i'll probably hit about 50% utilization this month. it's smarter and faster for my use case, and there's no wear and tear on the mac.

i'm not saying local llms are bad. they're just bad for my use case right now. the free tier feels nearly unlimited too. maybe if i was running models 24/7 it'd be worth it, but until my needs hit a certain level, it's not worth running local llms.