GPT-OSS

Why some small/medium models fail at grammar checking task?

Gem ma 4B, GPT-OSS-20B, and Qwen3-80B hallucinate spelling errors in grammatically correct sentences.