Hacker News new | ask | show | jobs
by roscas 314 days ago
Not a game dev but found out qwen3-coder:30b is a good balance for speed and quality for coding.

Almost don't use perplexity or chatgpt. When qwen3 fails with wrong answer, all others do the same.

But found out sometimes it is better than those online "services". Sure it is way better than gpt-oss:20b and that was a surprise.

Most of the times I just want to get sintaxe or code snippets.

It works also very well with long code and many programs from the same application.

So it runs on Ollama and I use OpenWebUI. It is very fast, way faster then gpt-oss:20b and it runs on old hardware AMD Ryzen 3600x with 10gb 3080 and 32gb of ram.

Never used ChatGPT 5 but the older failed too many times for me. Perplexity is ok, but I end up using search engines myself because the answers were not correct and many times very dangerous.