I’ve been testing the new focus search toggle in Kimi over the last week, mostly for work-related research. Coming from a background of handling a lot of Google Workspace and CRM migrations, I’m always wary of AI tools confidently presenting incorrect information, especially around technical specs or version histories.
My initial impression is that it does seem to ground the responses more firmly in the retrieved documents. For example, I asked it to compare the latest API changes between two SaaS platforms, and the answer directly cited release notes I had uploaded, with less “filler” speculation.
However, I’m curious how others are finding it:
* Does the feature noticeably slow down response times for you, or is it negligible?
* Have you encountered a scenario where it *still* hallucinated, even with focus search enabled?
* Do you use it for all queries now, or only for specific types of tasks?
I’m hoping this feature can become a reliable standard for reducing errors, especially when parsing migration guides or complex product documentation.
Migration is never smooth.
Your point about it citing release notes directly is spot on - I've seen the same when feeding it API docs. It cuts down on that vague, "in general, APIs often..." fluff that drives me nuts.
To your questions: the speed hit is definitely noticeable for me, maybe 2-3 seconds extra. I find myself only flipping it on for precise, document-based queries. For brainstorming or general explanation, I leave it off. It still seems to hallucinate on dates or version numbers if the source material is ambiguous, though it's much rarer.
Have you tried it with mixed document types? I'm curious if it prioritizes one format over another when searching.