alt Hacker NewsSingle shot or with reasoning enabled? My experience is that reasoning dramatically reduces hallucinations and improves output quality. I don't trust models without it.
Single shot or with reasoning enabled? My experience is that reasoning dramatically reduces hallucinations and improves output quality. I don't trust models without it.
> Overall, the best-performing model was Claude Opus 5 on “reasoning” mode, which still made mistakes in 39 per cent of answers.