I've been impressed by the quality of the Claude 3 lineup. If you don't need advanced math and reasoning, even Sonnet that is free reaches GPT-4 parity accordin

I've been impressed by the quality of the Claude 3 lineup. If you don't need advanced math and reasoning, even Sonnet that is free reaches GPT-4 parity according to the LMSYS Chatbot Arena Leaderboard: https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboar... There, Sonnet is within the margin of error of current GPT-4 and Opus the same but as for the GPT-4 previews. Above all Anthropic seem to have found themselves a very nice training set, watching how nicely results are retained as they go down in model sizes.