First Impressions
I ran the 33B model locally. It was surprisingly sharp. It solved a complex Python algorithm question that GPT-3.5 failed at. The fact that it's open source makes it incredibly attractive for privacy.
Best Feature: Performance/Cost
For an open model, it punches way above its weight class. It allows for offline coding assistants.
The Verdict
The hacker's choice. If you want to run your own AI, get DeepSeek.