I would still recommend trying gpt-oss-20b and seeing how it works for you, and also comparing it against other recent models around that size from other model series like Qwen 3 or (if you don’t need reasoning) Gemma 3.
Unfortunately, any model around that scale is going to have noticeable gaps in its knowledge of the world. Which model will work best—and whether any model will work well enough to be worth using—depends a lot on exactly what you want to accomplish, and there’s no substitute for trying out a few and deciding which one you prefer.
I would still recommend trying gpt-oss-20b and seeing how it works for you, and also comparing it against other recent models around that size from other model series like Qwen 3 or (if you don’t need reasoning) Gemma 3.
Unfortunately, any model around that scale is going to have noticeable gaps in its knowledge of the world. Which model will work best—and whether any model will work well enough to be worth using—depends a lot on exactly what you want to accomplish, and there’s no substitute for trying out a few and deciding which one you prefer.