An attempt to bring objectivity to the endless local vs API vs subscription debate
Press enter or click to view image in full size
The AI community is buzzing again.
Ollama just added Anthropic-compatible API support. GLM-4.7 Flash dropped with impressive benchmarks. My Twitter/LinkedIn feed is full of takes like “subscriptions are dead” and “local models for the win.”
Press enter or click to view image in full size
I get it. The dream of running frontier-quality AI on your own hardware, free from monthly fees and rate limits, is compelling. I want it to be true.
But I’ve been skeptical.
Not Medium member? Or want to share it with one? Read here!
My Reality Check
Last week I tried running GLM-4.7 Flash on my MacBook Pro(M4, 48GB memory) — a machine that costs over $3,000.
The results?
- It hung my machine during loading
- Once running, it managed ~16 tokens/second on a single task
- It consumed half my RAM