you may recall, but probably not, that GLM 5.2 was supposed to be the model that would kill the frontier labs roughly 4 weeks ago
For me it did. Cost/benefit of paying 20x for Claude/GPT over Codex+GLM-5.2 is minimal for real workloads I am running. For Kimi, I'm still not convinced it's as useful - same as I had for about every Deepseek release.
Painful fact: I need the bug-finding capabilities that they've removed from US models because 95% of my LLM use is code analysis. Imagine NOT finding an exploit in 3rd party software because the lab purposely hides it from you, then installing it, and then getting pwned. Nah breh... US labs are a liability.
For me it did. Cost/benefit of paying 20x for Claude/GPT over Codex+GLM-5.2 is minimal for real workloads I am running. For Kimi, I'm still not convinced it's as useful - same as I had for about every Deepseek release.
Painful fact: I need the bug-finding capabilities that they've removed from US models because 95% of my LLM use is code analysis. Imagine NOT finding an exploit in 3rd party software because the lab purposely hides it from you, then installing it, and then getting pwned. Nah breh... US labs are a liability.