DeepSeek opened an all-weekend off-peak lane less than a week after changing prices

The bottom line

Less than a week after changing V4 Pro pricing, DeepSeek has opened an all-weekend off-peak lane.

IT Home, Jiemian News and other outlets report that from 00:00 Beijing time on August 23, Saturday and Sunday are charged entirely at the off-peak rate. Weekdays keep the 09:00–12:00 and 14:00–18:00 peak windows. Charges incurred before August 23 remain under the old rules.

The change comes less than a week after peak pricing took effect on August 17. Reported V4 Pro output rates are RMB 27 per million tokens at peak and RMB 13.5 off-peak.

The original manuscript is most easily written as "DeepSeek's price increased fourfold, and the price butcher betrayed users." A more accurate version is: Some V4 Pro API prices are increased during peak hours, while offering off-peak scheduling discounts; it may still be cheaper than many leading-edge models, but low price is no longer an unconditional, 24/7 promise.

As of our editorial check, DeepSeek's public pricing page still mainly describes weekday peak windows and does not separately list the weekend exception. The timing and rule have been cross-reported by multiple outlets, but the exact model, account-balance scope and page update remain to be verified.

AI is adopting its own peak and off-peak pricing

DeepSeek’s change this time is not to simply increase the number, but to write the scarcity of compute directly into the price list: it is more expensive when everyone is using it, and it is cheaper when it can be offset.

This is especially important for batch processing, offline summarization, data cleaning, and nightly Agent tasks; for real-time chat, online code assistants, and peak production requests, prices may be more difficult to predict.

The official also emphasized that V4 Pro and V4 Flash support adjustable reasoning effort: low for simple tasks, high for daily agents, and max for complex tasks. In other words, model cost is now affected by three knobs at once: which model to pick, how deeply to think about it, and when to call it.

Price increase does not mean suddenly becoming more expensive

DeepSeek’s brand memory is “cheap”. So when they see the price change, many people’s first reaction is that it can’t hold on anymore.

But we can’t just look at the increase and not the absolute value. The increase in peak prices does not mean that it has surpassed all competitors; the 50% discount at low peaks does not mean that every user can easily move tasks to that period. What really counts is your monthly call volume, cache hit rate, failed retries, and wait time.

This is also a reversal of commercialization: the price butcher is not always free, but begins to write compute costs, demand peaks and service levels into the contract.

Less than a week later, weekends become the escape hatch

The practical value is not simply that Saturday and Sunday are cheaper. Developers now have a clear scheduling window for batch evaluation, offline summaries, data cleaning, embedding preparation and non-urgent agent jobs.

Do not use the weekend rate to budget real-time production traffic. Weekday peaks remain, and retries, longer context or higher reasoning effort can still raise the bill. Log request time, model, input/output tokens and cache hits before deciding whether the change actually saves money.

Our judgment: Cheap routes are entering the stage of refined operation

The price strategy of V4 Pro shows that DeepSeek not only wants to grab users at low prices, but also wants to divert traffic to more controllable time periods. For enterprises, this may be easier to budget for than a flat price increase; for individual developers, API bills and peak calls must be re-examined.

If you only call it occasionally, don't be scared by the word "price increase"; if you run a continuous agent, queue work for weekends where possible and separate weekday peak, off-peak, reasoning effort and cache-hit statistics.

Practical advice for developers

• Move deferrable jobs to the weekend, after confirming the rule applies to your account and model.
• Budget online production requests at weekday peak rates rather than the weekend discount.
• Reconcile charges before August 23 under the old rules, and keep logs and invoices for the first few days after the change.
• For V4 Flash, V4 Pro and other new models, use the account-level price page and invoice as the final source; do not generalize media reports to every model.

What to watch

• Whether the Peak/off-peak period will continue to be adjusted according to regions and products.
• Whether V4 Pro’s peak pricing impacts enterprise migrations and long-term contracts.
• Whether DeepSeek will extend off-peak charging to more models and products.
• When the official pricing page will explicitly document the weekend rule.
• Whether weekend off-peak usage brings different queueing, rate limits or service quality.

Frequently asked questions

#### Is the weekend rate half price all day?

According to the cross-reported notice, Saturday and Sunday are off-peak all day. Check your account invoice for the final applicable rate.

#### Did weekday peak pricing change?

Reports say the weekday schedule remains 09:00–12:00 and 14:00–18:00 Beijing time for peak hours, with the rest off-peak.

#### How are charges before August 23 handled?

They are settled under the previous pricing rules; the weekend change is not retroactive.

#### Does this apply to every model and account?

The public reports do not provide a complete scope. Treat this as unverified and check the price and billing page after signing in.