Deepseek Weekend Low Price Kaihe Aibox

At midnight on August 23, DeepSeek changed its pricing again. The rule is simple: Saturdays and Sundays no longer distinguish peak from off-peak — the lowest off-peak rate applies all weekend. For the flagship V4-Pro, output tokens that used to cost ¥27 per million at peak now cost ¥13.5 any time on a weekend — a straight 50% cut.

On the surface this looks like a "weekend discount," but for developers running agents, automation, and batch jobs — and for KAIHE AIBOX users — the signal matters more than whether prices went up or down. In one sentence: large models have started billing like electricity — when you use them now affects the bill more than which model you pick.

Two pricing moves in ten days

📖 Glossary

AI Box (also known as Agent Computer / Agent PC), is a dedicated local hardware device that runs AI Agents. Pre-installed with an AI agent management system, plug-and-play, running 24/7. Users can remotely command AI to work via Discord, Slack, Telegram, WhatsApp, and more.

This is already DeepSeek's second price change in ten days.

The first came on August 17: a broad price hike coupled with peak/off-peak metering. Peak hours are 9:00–12:00 and 14:00–18:00 Beijing time; everything else is off-peak, priced at exactly half. V4-Pro went from ¥6 per million output tokens before, to ¥27 at peak and ¥13.5 off-peak. The jump set the developer community on fire — some reported their operating costs at least tripled.

The second move is the August 23 patch: weekends are uniformly off-peak, no time-slicing. It tells you that grinding late Saturday night or suddenly calling the model Sunday afternoon both get the ¥13.5 floor rate.

Laid side by side, the trajectory is clear:

V4-Pro output: ¥6 before → ¥27 peak → ¥13.5 weekend/off-peak
V4-Flash output: ¥2 before → ¥9 peak → ¥4.5 weekend/off-peak

The real situation now: weekdays are expensive during the day, cheaper at night, and weekends are cheapest all day. It mirrors electricity exactly — peak rates during the day, valley rates at night. Model vendors have imported the power utility's playbook.

Why agent developers should care most

Casual users barely notice price shifts. Agents are different. An agent calls the model 24/7, over and over — batch data analysis, document processing, knowledge-base cleanup, scheduled crawlers, code review. These jobs easily burn millions or tens of millions of tokens.

The deciding factor is task type. A large share of what agents do does not require real-time response:

Summarize competitor prices and scrape industry news every dawn
Batch-clean last month's sales data over the weekend
Extract hundreds of PDFs into structured tables on a schedule
Run a full-site link health check deep at night

Shift these to the weekend or off-peak window and costs drop by half. An app burning 100 million output tokens a day pays ¥2,700 at V4-Pro peak, but ¥1,350 on a weekend or off-peak day. Run four weekends a month and that single line alone saves thousands.

It gets sharper when batch jobs stack with cache hits. DeepSeek's cached-input price goes as low as ¥0.15 (V4-Pro off-peak); if you repeatedly send the same system prompt and fixed knowledge base, input cost compresses further. Developers who schedule beat those who don't by up to 2x on the monthly bill.

KAIHE AIBOX: let the cheap window run for you, automatically

Metered model pricing means one new thing for ordinary users: you must choose not only which model, but when to call it. That collides exactly with what KAIHE AIBOX has always done — it is not another cloud model, but an always-plugged-in, always-on local base that schedules your jobs into the cheapest windows for you.

For KAIHE AIBOX users, this pricing shift is not a threat, it is a dividend. You keep one unit at home, drop the non-urgent batch jobs into it, and it stays online 24/7 with data never leaving your home. The backend orchestration engine automatically slots DeepSeek's work into the weekend low-price window. The model is just a tool being called — switching models or time slots is like swapping a battery. DeepSeek gets expensive, switch to Kimi; peak is pricey, queue it for the weekend. You never stare at a price table timing your moves.

Here is the point most people miss: no matter how cheap a cloud API is, it depends on your computer being on, the network being up, and the scheduler actually running. You sleep, the laptop locks, the cloud timer dies. KAIHE AIBOX's value is precisely that "the AI keeps working after your PC is off" — it decouples the agent's scheduling layer from your personal computer, so cheap windows are actually usable instead of stuck at "I have to babysit the PC to schedule."

To see how KAIHE AIBOX runs tasks while you sleep, read: AI Never Sleeps: KAIHE AIBOX Keeps Working After Your PC Is Off. If you already own one, KAIHE AIBOX-A1 is the most cost-effective entry model — a dedicated always-on executor for the home.

Collective price hikes make metered billing the norm

DeepSeek's "hike first, then patch in the weekend discount" is not isolated. In 2026 domestic models have already raised prices round after round: Alibaba Cloud, Tencent Cloud, Baidu Smart Cloud, and Zhipu AI all lifted API prices; Morgan Stanley calculated the average domestic output price rose 80% year over year in Q2. The industry price floor has clearly risen; "low-price customer acquisition" is shifting to "value pricing."

In other words, the future is not "which model is free," but "which model is cheapest when." Whoever can intelligently queue jobs into low-price windows and freely switch models without vendor lock-in holds the cost initiative.

KAIHE AIBOX has always been the agent-orchestration layer, not the inference layer — it does not compete on which model is smarter. It schedules the smartest model at the right time, at the cheapest price, with workflows stored at home and data never uploaded to the cloud. The harder the front-end models fight and the more prices diverge, the more valuable an always-awake local base becomes.

One line to take away

DeepSeek's weekend low-price halves V4-Pro output to ¥13.5 — at root, compute is now metered like electricity. For agent developers, queuing batch jobs into weekends and off-peak windows cuts costs in half. For KAIHE AIBOX users, this always-plugged-in local base automatically exploits the cheap windows — you never babysit the PC to time your moves; the model is just a tool, and orchestration is the real skill.

DeepSeek #WeekendLowPrice #LLMPriceHike #AgentCost #KAIHEAIBOX

Learn more: search 【KAIHE AIBOX】

Contact: [email protected]

—— KAIHE AIBOX · Your 24/7 Personal AI Assistant | AI Frontier