Choosing models made my budget look ugly again

Last Tuesday night, I was still revising a client proposal in my small studio in Hangzhou. The same project involved writing copy, making images, reviewing spreadsheets, and running some code. I had subscribed to several tools, and by the end of the month, the bill was approaching the cost of a part-time assistant. I wanted to move everything to “stronger” models, but I couldn't explain what each API call actually cost. I wasn't opposed to trying new tools; I was afraid the bill would suddenly jump.

What Microsoft actually released

In short, Microsoft released seven new models with more flexible tiers: inexpensive small models handle classification, summaries, and everyday drafts; more capable models handle complex reasoning, coding, and professional content. These aren't just for technical teams. Chen Yan, who runs a cross-border e-commerce business in Hangzhou, uses low-cost models to organize product information. She turns thousands of messy descriptions into a consistent format, then sends only the difficult cases to a higher-tier model. Instead of asking which model is best and using it for everything, she separates tasks first, then separates price points.

What it costs to replicate this setup

I did the math: starting with free trials or low-cost plans costs about 0 to 500 RMB. Understanding how the seven models map to my tasks takes roughly 2 to 4 hours. The technical barrier isn't high; web chat and copy-and-paste are enough to get started. The first step isn't researching every benchmark. It's opening Microsoft's model list and writing down the three things I do most often, such as “organizing customer messages,” “writing short-video scripts,” or “summarizing contract attachments.” Then I test one inexpensive model and one higher-tier model for each category. I got this wrong before too: I picked the most expensive model because it was trending, only to discover after the run that a small model handled plenty of the tasks.

Where I would start, depending on where I am

If I'm just getting started, I'll begin with the cheapest tier, get customer replies and document organization working, and hold off on long-term subscriptions. If I already have one or two clients, I'll create a small spreadsheet tracking the model used for each task, the cost per call, and the time spent fixing the output. If I'm scaling, I'll organize models into task tiers and set monthly budgets with automatic cutoff rules so higher traffic doesn't send costs out of control.

Not everyone needs all seven models. If most of my work still depends on manual conversations, I won't rush into comparing models. Once repetitive work starts piling up, trying the cheapest option will still be enough. There's no urgency to test them now. Delivering what I already have often costs less than chasing another new tool.