Ask HN: How does GPT-OSS compare to other open-source models? How does it compare to other open-source LLMs such as DeepSeek, Qwen, and Gemma, especially in terms of reasoning & coding ability? If you’ve tested it, did anything surprise you (good or bad)? Is it worth switching from an existing OSS model? Looking for real-world impressions, not just benchmarks. |
First comparison was a 500 line Python program that 5 minuts later, gpt-oss:20b was silent. Canceled.
Put the same program on qwen3-coder and in about 20 to 30 seconds it made the summary of what the program does. Just top!
But other examples were so bad, I did not removed yet so I can do a few more tests but I will remove it soon.
Qwen3.coder:30b is the best model I tested so far. Almost every prompt has output in one second or a little more.
Sometimes I put the same prompt in ChatGPT and Perplexity and almost everytime I get what I need from qwen3.
Since it is really fast and with quality output, that is almost my go to for help.