Scene hook
Last night I was still at my computer, staying up with a client to revise a proposal. The document kept getting longer, and the model kept getting more confused. I’ve been stuck here too: it’s not that I can’t write—once the material piles up, the responses start drifting off course.
What this is + who’s already using it
These past two days, I’ve been paying attention to a new category of models that can actually swallow ultra-long documents. Put simply, they can read more material in one go, and the expected cost is coming down, which feels friendlier for those of us doing consulting, content, and client delivery. The original article mentions that Moonshot AI released Kimi K3: a very large model that can handle extremely long context, with plans to open the weights. I read that as: in the future, we may not only use it on someone else’s website—we may also be able to deploy it ourselves, or through a service provider. Why do I care? Because last Tuesday at 9 p.m., my friend Lin Wei was in a coworking office in Xuhui, Shanghai, replying to a client on WeChat while pasting in four interview transcripts for the model. The result contradicted itself from one answer to the next, and she immediately wanted to switch approaches. I got this wrong before too—I used to think a smarter model was enough. Later I realized that, a lot of the time, the real problem is that it never fully took in the long material.
What it costs to try today
If I just want to test it today, the cost looks like this: money: RMB 0-50; time: 30-60 minutes; technical barrier: no coding needed. I’d start with the official chat version or a third-party integration, and skip the complicated self-hosting steps for now. First step: open the product homepage and click something like “Start Chat” or “New Project,” then drop in one real long document—client requirements, a podcast transcript, or a course outline—and ask only one question first: “Please summarize this first; don’t jump into solutions yet.” Not everyone needs this tool. If I’m not dealing with long documents or complex source material yet, it’s also fine not to try it now.
Advice by stage: just starting / 1-2 clients / scaling up
If I’m just starting out, I’d use it first for “organizing materials.” I wouldn’t rush into fully automated drafts; I’d just try to save myself 30 minutes of repetitive reading.
If I already have 1-2 clients, I’d place it in the step before revising a proposal: feed it past chats, requirement docs, and pricing notes first, then ask it to find conflicts.
If I’m scaling up, I’d focus on two things: whether the price can come down, and whether my materials can avoid being locked into a single platform. What small teams like ours fear most isn’t that a tool isn’t new enough—it’s becoming more dependent on it over time, until the monthly cost and the risk both go up together.