I totally missed this rule

Last week I was rushing a client proposal and didn't wrap up until past midnight — used AI to polish a draft. The next day when I checked the bill, I realized the exact same tool had charged me nearly double for daytime vs. nighttime runs. I just sat there staring at the screen. I'd basically thrown away a dozen-plus yuan for nothing.

So what does DeepSeek's new rule actually say

DeepSeek (the Chinese large-model company) just rolled out time-of-use pricing. Plain English: 8 AM to midnight is "peak," full price; midnight to 8 AM is "off-peak," roughly 50% off.

My friend Xiaolin, who runs a Xiaohongshu (Little Red Book) account matrix out of Hangzhou, only kicks off her AI batch-copy generation after 11 PM. Her line: "All the AI I queue up during the day is paying full price." Move it to nighttime and she saves a coffee's worth a month.

Then there's Lao Zhou — indie dev in Shenzhen, building an accounting tool. He set up a Feishu (Lark) bot that bundles every pending task at 10:30 PM and ships them to AI, then he grabs the results at 7 AM when he wakes up. "Cost cut in half," he told me.

What it costs you to start using this today

Money: If you call DeepSeek's API directly (plain English: let your program talk to DeepSeek automatically), off-peak runs at roughly half the regular rate. A few dozen overnight jobs a month saves you 20–30 yuan.

Time: Regular users — zero steps. If the app or web version you already use runs on DeepSeek under the hood, it auto-applies the discount during off-peak hours.

Technical barrier: Just manually shifting when you use it. No tech skills needed. Want it automated? You'll need a little scheduled-task knowledge — but honestly, just ask your tool vendor to flip it on for you.

First step: Check if the AI tools you already use run on DeepSeek (more than half of domestic Chinese AIs do). Then move all your "non-urgent" tasks — outlines, copy edits, translation — to after 11 PM.

How to actually use this rule at your stage

Just starting out, tight on budget: push every "non-urgent" AI task — outlines, copy edits, translation — to nighttime. A month of that saves you a milk tea.

You have 1–2 stable clients: AI proofreading and polishing before deliverables — run it at night, save money, don't delay the work. Your clients will never know which time window you used.

Scaling up: consider connecting DeepSeek's API (have your tool vendor enable it — no coding on your end). Queue up batch jobs for off-peak. A month in, your AI line item could literally halve.

This rule isn't for everyone — if you only chat with AI for a couple minutes a day, the savings are basically nothing. But if AI does more than 10 minutes of work for you daily, try it tonight.