Claude 3.7 Sonnet: A One-Hour Review
Claude 3.7 just dropped, and I tried it immediately.
For one, coding is definitely better. Compared to 3.5 Sonnet, agentic coding feels much more natural—it writes its own test code, and if tests fail, it fixes them and reruns. I can count on one hand the times I had to step in.
That said, Korean responses still feel a bit awkward. It has improved a lot, but occasionally I see mixed speech levels or verb endings that just fall off oddly. Still, at this level, I think I'll go with this for the next project instead of Code Llama.
10 answers
Agreed, I've used it too and it was good.
Is agentic coding really that good? I didn't notice much of a difference compared to 3.5.
It's a bit of a shame that the Korean is awkward. I get weird verb endings sometimes when working on long documents too lol.
Oh, I didn't know that.
When coding, it was really amazing to watch it write and fix its own test code. I never intervened even once.
Do you have a source?
Well, isn't that a bit much? I'd still be uneasy using it in production. Especially for Korean.
This is right lol, but Korean answers still have a lot of room for improvement.
Awkward Korean can be solved to some extent with system prompts. I added 'always use honorifics, natural endings' and it got much better.
I felt the same way. The sentence endings in Korean responses are often inconsistent, but overall I'm satisfied.