I know people just like to throw shit at Anthropic. I'm not one of those. I have nothing against them as a company, and I actually dislike them less than the other big players. I had all my team switch over from Cursor because Opus felt so good. Since the Max plan is never enough, expenses are growing bigger by the day. So when we can we supplement with Qwen 3.6 plus keeping Opus as harness. It's good, but wasn't "as" good. Lots of mistakes and stubs.
The feeling everyone is sharing is Opus 4.7 got suddenly so lazy, on top of expensive. Part of the problem might be in Claude Code CLI itself, who knows.
And so today I switched over to kimi 2.6 and it's.. wow! So fast and pleasurable to use. Context is much smaller but keeping an eye on it it's still pretty reliable. Claude is happy going back and forth with questions and spammy tool outputs.. seems the Kimi team worked to manage their smaller context better perhaps? More testing is needed to say this for certain. But I immediately purchased a yearly subscription and will recommend to my colleagues as well.
At the moment I'm using it with their cli, it feels smoother than it is when plugging it into CC via env vars. I'm just a bit sad it doesn't work out of the box with Forge. I submitted a PR to fix it (https://github.com/tailcallhq/forgecode/pull/3098).
[link] [comments]