· ·Sources
DeepSeek charges full price for only 35 hours a week — every other hour, including all weekend, is half price. Here are those peak windows in your own timezone, the exact rates for V4 Flash and Pro, and — the part nobody tells you — whether the discount actually beats just using a cheaper provider.
DeepSeek defines two short peak windows in UTC and discounts everything else, including the whole weekend. Pick a timezone to see them converted — then hold them against your own traffic curve, because that comparison is where the real answer lives.
Peak windows taken from DeepSeek's official pricing page, checked September 2026. Off-peak covers 133 of 168 hours a week — but that figure assumes your traffic is spread evenly across the day. Real traffic is not, and for some time zones these windows land squarely inside working hours.
Leave your traffic alone and it lands off-peak about 79% of the time — that is the baseline below, and it costs you nothing to get. Slide it up if you batch work into quiet hours, or down if your busy hours collide with the peak windows above.
* DeepSeek Official row shows your blended rate: peak and off-peak weighted by the slider above.
* OpenRouter rows are the cheapest host at time of check: deepseek-v4-flash-0731 / deepseek-v4-pro-0813.
The undated IDs route differently and can be pricier — on Pro, deepseek-v4-pro costs more than DeepSeek's own off-peak rate. Always treat a router price as a floor, not a quote.
* No provider here pays us to be ranked first — the cheapest row is highlighted whatever the commercial arrangement.
* DeepSeek Official "average" here is the simple peak/off-peak midpoint ($0.33/$0.99 for Flash). Weight it by the clock instead and the expected rate falls to about $0.266/$0.798, because 79% of the week is off-peak. Which figure applies to you depends on when your traffic actually runs: the midpoint is the cautious one, the time-weighted one assumes you are spread evenly.
* Cached input at DeepSeek Official is far cheaper again (Flash $0.007–0.014/M, Pro $0.022–0.044/M). If your prompts repeat, that dwarfs the off-peak discount.
Off-peak pricing is demand shaping. DeepSeek is buying flexibility: if you can move, you get half price; if you cannot, you pay for the privilege of peak capacity. That is a fair trade, but it is worth being clear about what you are actually getting.
Catch one — the discount is off a high base. Halving $0.44 still leaves $0.22, which is more than OpenRouter charges at any hour. The percentage sounds dramatic; the absolute number is what hits your invoice.
Catch two — 79% is generous, but only if your traffic is spread out. The peak windows sit at UTC 01:00–04:00 and 06:00–10:00 on weekdays. For a UTC+8 user that is 09:00–12:00 and 14:00–18:00 — the core of the working day, twice over. For a UTC−4 user it is 21:00–00:00 and 02:00–06:00, where only the evening rush collides. Same pricing, opposite exposure: if your users are awake during your local daytime and you sit east of UTC, a lot of your traffic can land in peak hours. Check your own case with the converter above — 79% assumes uniform traffic, and real traffic is nothing like uniform.
Catch three — on a routing marketplace the model ID you call decides the answer. OpenRouter lists DeepSeek V4 Pro twice: deepseek-v4-pro-0813 (cheapest host Alibaba, $0.58/$1.74) and deepseek-v4-pro (cheapest host DigitalOcean, $0.87/$1.74). At our default volume the first beats DeepSeek's off-peak rate by 12% — and the second loses to it by 13%. Same model family, same marketplace, opposite verdict. Flash is not sensitive this way: every route we checked undercuts off-peak Flash. If a comparison quotes you a single OpenRouter number without the model ID, it is not telling you enough to act on.
Our honest read: the discount is far broader than most people assume — 79% of the week, weekends included, applied with no opt-in. If you call DeepSeek's own endpoint you are probably already getting it, and there is no scheduler to build. That still does not make DeepSeek cheapest. On Flash, every alternative we checked — Novita, and both OpenRouter model IDs — undercuts off-peak Flash at every hour of the day, so scheduling effort is better spent elsewhere. On Pro the answer splits: DeepSeek's off-peak rate halves your official bill and badly beats Novita, and it beats the pro-0813 route by about 12% at Alibaba's price — but it loses to the plain deepseek-v4-pro route by about 13%. The discount makes the official endpoint reasonable; it does not make it the cheapest option. Run your own numbers above rather than trusting the rule of thumb — the break-even point moves with your input/output ratio and with how much of your traffic lands in those 35 peak hours.
Here is the finding we did not expect when we started checking. OpenRouter's deepseek-v4-pro-0813 page lists DeepSeek itself as one of the hosts — at $0.66 / $1.98, which is exactly DeepSeek's own off-peak rate. It does not move with the clock.
Route through that host and the peak window stops mattering: you pay the off-peak price at 08:00 UTC on a Wednesday, which is when a direct call to api.deepseek.com would cost you $1.32 / $3.96 — double. At our default volume that is $115.50 instead of $231.00 for identical traffic through an identical model, and the only thing you give up is DeepSeek's direct billing relationship.
That reframes the whole question. If you want DeepSeek's own model at DeepSeek's best price, routing is a one-line change and scheduling is unnecessary. Alibaba's $0.5808 / $1.7424 on the same model ID is still marginally cheaper, so the router wins on price either way — but the point is that you no longer need to care what time it is.
Every hour that is not peak. Peak is UTC 01:00–04:00 and 06:00–10:00, Monday to Friday — 35 hours out of 168. That leaves 133 off-peak hours, including all 48 weekend hours, plus weekday 00:00–01:00, 04:00–06:00 and 10:00–24:00. Use the converter above to see the two peak windows in your own timezone.
Yes — all weekend, Saturday and Sunday, every hour. Peak hours are defined Monday to Friday only.
Half price — off-peak is 50% off the peak rate. Flash goes from $0.44/$1.32 to $0.22/$0.66 per million input/output tokens; Pro goes from $1.32/$3.96 to $0.66/$1.98.
For Flash, no — every route we found beats off-peak Flash ($0.05/$0.16 on deepseek-v4-flash-0731, $0.068/$0.168 on deepseek-v4-flash) by a wide margin.
For Pro, it depends which model ID you call, and the two disagree. deepseek-v4-pro-0813 routes from $0.58/$1.74, below off-peak Pro's $0.66/$1.98. deepseek-v4-pro routes from $0.87/$1.74, which is above it. OpenRouter hosts shift, so re-check before relying on any of these.
Effectively yes, via a router. OpenRouter lists DeepSeek's own endpoint as a host of deepseek-v4-pro-0813 at $0.66/$1.98 — DeepSeek's off-peak rate, charged around the clock. Calling api.deepseek.com directly during peak hours costs $1.32/$3.96 for the same model.
The discount is applied per request based on when the request arrives, in UTC — there is no plan to switch to. Confirm the current mechanism on DeepSeek's pricing page before designing around it.
Cached input is a separate and much steeper discount ($0.007–$0.044/M). Whether the two stack is a detail you should confirm on the official pricing page rather than assume — either way, cache reuse is the larger lever of the two.
For Flash, yes: Novita charges $0.14/$0.28 around the clock, below DeepSeek's off-peak $0.22/$0.66. For Pro, no — Novita is $1.60/$3.20, well above DeepSeek's off-peak $0.66/$1.98.
One caveat worth knowing: Novita also lists dated builds — deepseek-v4-flash-0731 at $0.44/$1.32 and deepseek-v4-pro-0813 at $1.32/$3.96, which are DeepSeek's peak rates. The cheaper undated entries are the ones we compare against above. Which one you want depends on whether you need the dated re-post-trained build.
Every figure on this page is traceable to one of these. Peak windows and all DeepSeek rates come from DeepSeek's own pricing page — quoted below so you can check us:
Official wording, DeepSeek pricing page footnote (1):
“Off-peak rates are half of the peak rates. Peak hours are 01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday (all other hours are off-peak).”
We price everything in USD and quote DeepSeek's international rates. DeepSeek also publishes yuan pricing for mainland China — a separate page, a separate currency, deliberately not used here.
Peak windows taken from DeepSeek's official pricing page, September 2026. Rates checked August 2026 against the same page and verified against live dashboards where available. Windows, discount depth and reseller routing all change — verify current rates before committing spend.
Transparency: links marked ref link are affiliate links — we may earn a commission if you sign up, at no extra cost to you. This never affects the comparison: the cheapest option is highlighted in green regardless of who pays us. Verify current rates before committing spend.