I experienced Kimi K3 Max in OpenCode The OpenCode GO package I subscribed to last month for GLM 5.2 (first month $5) will expire tomorrow. But there are still some unused quotas. Today, Kimi K3 arrived. I haven't subscribed to the official ¥ 99 package yet, who knew I could use Kimi K3 when I opened OpenCode. Hurry up and try it out to see how the experience is. And the result? The first feeling is: Slow down. If you feel slow with the previous GLM 5.2, it's best to be mentally prepared. At least in the OpenCode GO package, using Kimi K3 guarantees to exhaust all your patience. So much so that I clicked on the Thinking steps of the model to carefully examine it, because I was worried that it might be stuck and not moving. The second feeling is: I've been thinking for too long. These two should be related because thinking too much leads to a very slow physical sensation. Could it also be an issue with OpenCode itself? What is the intelligence level of Kimi K3 Max? From its comprehensive analysis and interpretation of the current project, the problems pointed out, and the improvement plans provided, it is indeed good. The webpage project being modified is one that Grok 4.5 has been repeatedly working on, but there are still many issues that need to be resolved. Now Kimi K3 has taken over, pointing out a bunch of problems all at once, and is now making improvements with great enthusiasm. I am more confident this time. Speaking of Kimi (K2 era) back then, it gave me the feeling of being like a bean bun, making me unable to feel the desire to try (because the name doesn't feel too formal). Unexpectedly, in less than a year, Kimi came with his K3 with 2.5 trillion parameters. Given that the intelligence level of the model is already high enough, the actual user experience (speed, cost) becomes more important. I hope Kimi can make improvements in this area as soon as possible. If the speed is fast and the price is affordable, it may really become a substitute for GPT 5.6 and Claude Opus 4.8.