Claude, QWEN, and Perplexity have been the best performing imo. And what I like about Perplexity is it gives a list of sources it used to generate the response, which is helpful to further fact-check the response.