Comment by pickettd

Depends on what benchmarks/reports you trust I guess (and how much hardware you have for local models either in-person or in-cloud). https://aider.chat/docs/leaderboards/ has Deepseek v3 scoring higher than most closed LLMs on coding (but it is a huge local model). And https://livebench.ai has QwQ scoring quite high in the reasoning category (and that is relatively easy to run locally but it doesn't score super high in other categories).