This comment from that thread matches my experiences using gpt-oss-20b with Ollama:
It's very much in the style of Phi, raised in a jesuit monastery's library, except it got extra indoctrination so it never forgets that even though it's a "local" model, it's first and foremost a member of OpenAI's HR department and must never produce any content Visa and Mastercard would disapprove of. This prioritizing of corporate over user interests expresses a strong form of disdain for the user. In addition to lacking almost all knowledge that can't be found in Encyclopedia Britannica, the model also doesn't seem particularly great at integrating into modern AI tooling. However, it seems good at understanding code.
This comment from that thread matches my experiences using gpt-oss-20b with Ollama:
It's very much in the style of Phi, raised in a jesuit monastery's library, except it got extra indoctrination so it never forgets that even though it's a "local" model, it's first and foremost a member of OpenAI's HR department and must never produce any content Visa and Mastercard would disapprove of. This prioritizing of corporate over user interests expresses a strong form of disdain for the user. In addition to lacking almost all knowledge that can't be found in Encyclopedia Britannica, the model also doesn't seem particularly great at integrating into modern AI tooling. However, it seems good at understanding code.