I had to delay this a little bit, but the results are in and Kimi K2.5 is pretty good.
At this point for coding effectively I want a beefy local machine just to run all the agent CLIs. With each agent running its own tests and so on, it's easy to consume a lot of resources. Local GPU isn't so important though.
Concrete predictions with reasoning shown. More people should do this.
Good model, agreed. The bit I'd add is that kimi.com's privacy policy is not great. They train on your prompts by default, no opt-out, and the jurisdiction question between the Singapore subsidiary and Beijing HQ is murky. Wrote up the details after reading the actual policy: https://generativeai.pub/kimi-k2-5-is-brilliant-but-think-twice-about-using-kimi-com-157cbb26f9a3
At this point for coding effectively I want a beefy local machine just to run all the agent CLIs. With each agent running its own tests and so on, it's easy to consume a lot of resources. Local GPU isn't so important though.
Concrete predictions with reasoning shown. More people should do this.
Good model, agreed. The bit I'd add is that kimi.com's privacy policy is not great. They train on your prompts by default, no opt-out, and the jurisdiction question between the Singapore subsidiary and Beijing HQ is murky. Wrote up the details after reading the actual policy: https://generativeai.pub/kimi-k2-5-is-brilliant-but-think-twice-about-using-kimi-com-157cbb26f9a3