I think most people who've tried them both would tell you Anthropic's models are more than marginally better than Kimi. Kimi and the other open source models may score well on SWE-bench or whatever but the gap is noticeable IMHO once you actually try to use them.

It depends on what your task is and how precise your prompts are. Planning with fable or 4.8 and laying out the plan in step by step process and coding with mimo v2.5 pro or dsv4pro or qwen 3.7 max and doing a final review with 5.5 has worked really well for me for infra stuff.