I do not have direct experience with Kimi Code K2.7 because running it locally requires insane resources. Even at INT4 quantization it demands nearly 580 GB of VRAM which means an enterprise 8x H100 node.

I prefer hosting my LLMs locally on manageable hardware which is why I stick with Qwen 3.6 currently.

However if you look at the Arena Agent Leaderboard Kimi K2.7 Code is currently the top performer among open weight options for agentic workflows sitting right behind the absolute latest closed source frontier models.
https://arena.ai/leaderboard/agent

Let us know l, if you have real live test results.
#asknostr