Sonnet 5.0 vs 4.6
用户对 Sonnet 5.0 和 4.6 进行了对比测试，发现 4.6 在遵循指令方面表现更好，而 5.0 则试图为自己辩护。尽管 5.0 认为单个案例不能代表趋势，但用户在创意写作中多次发现 5.0 的表现远不如预期，Opus 也存在类似情况。用户建议其他人可以自行测试，并预测 Sonnet 4.6 的表现通常会更好。…
I asked both Sonnet 5.0 and 4.6 the same questions about their comparative performance, and then talked to each model directly. You can see for yourself the complete loss of performance between 4.6, which reasons for you according to your instructions, and 5.0, which tries to defend itself, user be damned.
As mentioned by 5.0 at the end in its defense, one case here is indeed not a trend by itself, but I've experienced so many cases in creative writing I've already noticed 5.0 greatly underperforming. A similar trend exists with Opus.
It's an interesting exercise you can repeat on your own accord with different prompts. Nine times out of ten, you should find Sonnet 4.6 does better.
These are the last days of creative writing on Claude. GPT died around last year when everything after 4.5 censors to such an absurd extent a pedestrian PG story will get censored in less than 8 hours of narrative time. Claude creative writing will die once 4.6 is removed. Enjoy it while it lasts.