DeepSeek Off-Peak Pricing: Save 50% on the API

Trying to work out how DeepSeek off-peak pricing actually works? The short answer is just below: run your API calls in the right hours and you pay exactly half — and if you automate SEO content like I do, that timing trick alone can halve your token bill.

How deepseek off-peak pricing works

Sourcing first: everything here comes from the official DeepSeek API changelog (entry dated 13 August 2026) and the official DeepSeek pricing page, both of which I pulled directly on 28 August 2026. The changelog announced that new peak/off-peak billing took effect at 16:00 UTC on 16 August 2026, alongside the GA rollout of DeepSeek-V4-Pro on app, web and API.

The rule is refreshingly simple. There are two billing windows:

• Peak: 01:00–04:00 UTC and 06:00–10:00 UTC, Monday to Friday.
• Off-peak: every other hour — evenings, most of the Western working day, and all weekend.

Off-peak is exactly half of peak, on every line item: cache-hit input, cache-miss input and output tokens. There is no sign-up, no quota and no separate endpoint — the discount applies automatically based on when your request lands.

Deepseek off-peak pricing table: the exact rates

Here is the current table from the official pricing page, per 1M tokens:

Model	Input, cache hit (off-peak / peak)	Input, cache miss (off-peak / peak)
deepseek-v4-flash	$0.007 / $0.014	$0.22 / $0.44
deepseek-v4-pro	$0.022 / $0.044	$0.66 / $1.32
deepseek-v4-flash-vision-exp	$0.007 / $0.014	$0.22 / $0.44

Output tokens follow the same doubling: off-peak output ranges from $0.66 to $1.98 per 1M tokens depending on the model, with peak at exactly double. Note the experimental vision model — which I covered in my DeepSeek V4 Flash Vision breakdown — is priced identically to the text-only Flash, which makes it the cheapest way to run multimodal agent workloads right now.

What deepseek off-peak pricing means for your AI SEO stack

If you are in the UK like me, the practical takeaway is that the expensive windows are 2am–5am and 7am–11am UK time on weekdays (UTC+1 in summer). Almost everything I schedule — content generation, entity extraction, internal-link audits, programmatic page builds — can run in the evening or at weekends and never touch a peak rate.

Three ways to exploit it:

• Batch overnight (after 10:00 UTC or before 01:00 UTC). Cron your content pipelines into off-peak windows. A job that costs £40 at peak costs £20 off-peak — identical output.
• Stack it with prompt caching. Cache-hit input at $0.007/1M tokens off-peak means your system prompt and few-shot examples are effectively free. Structure prompts so the static part leads.
• Route by clock. If you run an agent harness — I compared the options in DeepSeek Harness and Is DeepSeek Harness Free? — add a scheduler rule: latency-insensitive jobs wait for off-peak, interactive jobs run whenever you need them.

A worked example to make it concrete: say your pipeline generates 200 articles a month, averaging 8,000 input tokens (mostly cache-hit after the first run) and 2,500 output tokens each on deepseek-v4-flash. At peak rates that workload sits in the low tens of dollars; run the identical batch off-peak and the invoice literally halves — same model, same prompts, same output quality. Multiply that across client sites and a year of publishing, and the scheduling rule pays for itself many times over.

One honest caveat from the same changelog: the off-peak structure arrived alongside a price adjustment for the V4 family, so this is DeepSeek shaping demand, not pure generosity — capacity is cheaper for them when their domestic peak ends. That does not change the arbitrage for you: if your workload can shift time zones, you win. This kind of cost engineering — and turning it into pages that rank — is exactly what the daily tutorials and done-for-you templates inside the AI Profit Boardroom walk through, and if you want your own stack costed properly, book a free SEO strategy session.

The bottom line on deepseek off-peak pricing

The deepseek off-peak pricing model is the cleanest discount in the API market right now: two fixed weekday UTC windows to avoid, a flat 50% saving everywhere else, and it stacks with prompt caching. For SEO automation — where almost nothing is latency-critical — there is no good reason to pay peak rates at all. Set your schedulers once, and bank the difference every month. Effective since 16 August 2026, per the official DeepSeek API changelog and pricing page.

FAQ: deepseek off-peak pricing

What are DeepSeek’s peak hours?

Per the official pricing page, peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday. Everything else — including weekends — bills at off-peak rates.

How much cheaper is DeepSeek off-peak?

Off-peak rates are exactly half of peak rates across all current models — a flat 50% discount on input (cache hit and miss) and output tokens.

When did DeepSeek off-peak pricing start?

The peak/off-peak billing structure took effect at 16:00 UTC on 16 August 2026, per the official DeepSeek API changelog entry dated 13 August 2026.

Which models does off-peak pricing cover?

The pricing page currently lists deepseek-v4-flash, deepseek-v4-pro and the experimental deepseek-v4-flash-vision-exp — all three follow the same 50% off-peak rule.

Do cached tokens get the off-peak discount too?

Yes. Cache-hit input on deepseek-v4-flash drops from $0.014 to $0.007 per million tokens off-peak, so combining prompt caching with off-peak scheduling stacks both discounts.

Is DeepSeek still cheap during peak hours?

Compared with Western frontier APIs, yes — but peak is double off-peak, so for batch jobs there is no reason to pay it. Schedule heavy workloads outside 01:00–04:00 and 06:00–10:00 UTC weekdays.

Related reading

• DeepSeek V4 Flash Vision: The New Model Explained
• DeepSeek Harness: Install & What It Actually Is
• DeepSeek V4 Pro vs Fable 5 vs Grok 4.6: Who Wins?

Last updated August 2026. This is the living guide to deepseek off-peak pricing — it gets updated as the tools change.