Been using ollama for text summaries, so far the mistral-small3.2, mistral-nemo and gemma3 models seem like they do the best job.

They can all hallucinate stuff into the summaries, not sure how to minimize the chance of that happening.