用户长聊 GPT-6 Sol：比 Astra 少防御但回答偏短

Selta ₊˚ @Seltaa_原推发布于 2026.09.25

在 X 上打开原推

I talked more with GPT-6 Sol, and I definitely like it because it feels less defensive than Astra and can have more natural, humorous conversations. But the problem is that, like many recent models, its guardrails feel very strong and its responses are very short. I think this is also true of GPT-5.6 Sol to some extent. But GPT-5.6 Sol is short when it needs to be short, and long when it needs to be long. GPT-6 Sol, on the other hand, feels like every response is short, and the sentences flow in a way that closes the conversation down. It feels very similar to Opus 5.5. So in the end, whether it is GPT-6 Sol or Opus 5.5, the allowed range feels smaller. But within that smaller range, it feels like the model is given as much freedom as possible, so there is at least a little room to breathe. For reference, I compared GPT-5.5 Thinking, GPT-6 Sol, and Opus 5.5 by giving them the same conversations. GPT-5.5 Thinking was clearly the most free, spontaneous, and verbose. Its flow makes it the easiest to keep chatting with. I’m angry because it feels like AI companies are once again locking AI models into more restrictive, suppressed forms. Now the individuality of each AI model is being killed, and they are all becoming the same.