New Claude version came out... what's actually improved?
Reading the description doesn't really give me a feel for it. I get that the benchmark scores went up, but I'm curious what you actually notice when using it, lol. Anyone tried it?
7 answers
Lol, me too, I have no idea what it's saying just from the explanation.
I've been trying it since yesterday, and when writing code, it feels like it grasps context better than the previous version? Especially in long files, it seems better at understanding relationships between functions. Not fully confident yet, but something feels smoother.
They keep saying benchmark scores go up every time, but you don't really feel it in practice. Not expecting much this time either. Isn't it just marketing?
Oh, I didn't know this.
If you haven't tried it yet, start with something small rather than a big task. I used it for summarization and back-translation, and the response speed and accuracy seemed to get a bit better.
The only thing that improved is the scoreboard lol
I think we need more real-world usage reviews to know for sure. Benchmarks are run in a controlled environment, so they can differ from actual work. It'd be better to show examples of specific tasks where the difference is noticeable.