​
I have an Ollama Pro subscription and use the cloud LLM versions for OpenClaw.
I had Kimi 2.5 for two months right from the start, and it was pretty suboptimal. Then I switched to GLM 5.1—it was amazing what a difference that made, and I finally had an agent that did everything right on its own and didn’t forget a thing.
But now to the problem:
GLM uses up too much computing power. I used up 22% of my weekly usage in a single day. I won’t make it through the whole week at this rate.
So I switched to Deep Seek Flash for two days. All I can say is, what an idiot… It ignores clear instructions, doesn’t read what’s in the tools, memory, or self-improvement sections, and just guesses!
Now I’ve been on Kimi k2.6 for a few hours. He’s 50/50 stupid, and I have to constantly correct him. He also doesn’t read what’s in the Skill.md, etc. But he’s better than Deep Seek.
Glm was the holy grail for me; he did everything right, and I didn’t have to explain, correct, or criticize anything. With these two, I’m constantly tweaking every task he has.