Race to the bottom?
这条记录涉及编程工具或代码能力更新，适合开发者评估工作流变化和可复用价值。
As far as consumer uses and vibe coding, I have great success with yesterday’s models. Claude Code Sonnet 4.5 produces great code with nearly no errors. I tried Fable for my work, and got no improvement, just more cost. I’m talking consumers here. Enterprises with large code bases may need larger models just to hold the bigger contexts, but I’m seeing they probably dont either once they have the right processes in place.
Sure there are some tasks that require more juice, protein folding, chemistry, etc. But for the vast majority of users and most solo vibe-coders the value is flattening out fast.
Give me fast, low cost models from ‘yesterday’ and with a good process, you can stop wasting all that electricity and money for 95% of all the users who dink around with AI as a better Google or a reliable way to build their own stuff.