How much did DeepSeek V4 prices increase in August 2026?

DeepSeek V4-Pro cache-miss input rose from $0.435 to $0.66 off-peak and $1.32 at peak per million tokens, with output rising from $0.87 to $1.98 off-peak and $3.96 at peak. V4-Flash rose from $0.14/$0.28 to $0.22/$0.66 off-peak and $0.44/$1.32 at peak. Cache-hit input saw the largest increase — up to roughly 1,100% on V4-Pro at peak, from $0.003625 to $0.044. The new pricing took effect at 16:00 UTC on August 16, 2026.

What are DeepSeek's peak hours?

Peak hours are 01:00-04:00 and 06:00-10:00 UTC — seven hours per day. All other 17 hours are off-peak and billed at exactly half the peak rate. This schedule falls during Chinese business hours, so North American and European users doing daytime work are largely billed at off-peak rates.

Is DeepSeek still the cheapest AI API after the increase?

Off-peak, yes — V4-Flash at $0.22/$0.66 costs about $0.0099 on a 30K-input/5K-output task versus $0.012 for GPT-5.6 Luna. At peak, V4-Flash costs $0.0198 and loses to Luna. The decisive change is the cache-hit rate: DeepSeek's roughly 98% caching discount was the mechanism behind its cost-per-task lead, and that is what the new schedule repriced hardest.

Quick Answer

DeepSeek V4 Price Increase Aug 16, 2026: New Rates

Published: August 17, 2026

The Short Answer

On August 16, 2026 at 16:00 UTC, DeepSeek raised API prices across the V4 family and introduced peak/off-peak rates for the first time. Headline increases run from 51% to roughly 1,100%, depending on the token type and hour.

DeepSeek is still cheap. It is no longer categorically cheap — and the part that got repriced hardest is exactly the part that made it cheapest.

The New Pricing

All figures per million tokens, effective August 16, 2026:

	V4-Flash off-peak	V4-Flash peak	V4-Pro off-peak	V4-Pro peak
Input (cache hit)	$0.007	$0.014	$0.022	$0.044
Input (cache miss)	$0.22	$0.44	$0.66	$1.32
Output	$0.66	$1.32	$1.98	$3.96

What it replaced: V4-Flash was a flat $0.14 / $0.28 (cache hit $0.0028). V4-Pro was a flat $0.435 / $0.87 (cache hit $0.003625).

Peak hours are 01:00-04:00 and 06:00-10:00 UTC — seven hours. The other 17 hours are off-peak at exactly half the peak rate.

Both models carry a 1M context window and a 384K max output. Concurrency limits are 2,500 for Flash and 500 for Pro.

The Increase, By Token Type

Token type	V4-Pro increase	V4-Flash increase
Input (cache miss)	+51% off-peak → +203% peak	+57% → +214%
Output	+127% → +355%	+136% → +371%
Input (cache hit)	+507% → +1,113%	+150% → +400%

The cache-hit row is the story. DeepSeek’s cache discount was roughly 98% off input, against an industry norm nearer 90%. That single mechanism is what held DeepSeek’s measured cost-per-task well below rivals even after OpenAI cut GPT-5.6 Luna by 80% on July 30, 2026. The new schedule targets it directly: V4-Pro cache hits went from $0.003625 to $0.044 at peak — a 12x increase.

What Else Shipped

The price change was announced inside the V4-Pro general availability post on August 13, 2026, not as a standalone notice. Alongside it:

V4-Pro is GA; V4-Flash is in beta.
Flexible reasoning effort on both models: low for simple tasks, high for daily agent workflows, max for complex work.
Native OpenAI Responses API support, with one-click Codex setup.
An Anthropic-format endpoint at api.deepseek.com/anthropic alongside the OpenAI-format base URL.
V4-Pro is live in the app and on the web under “Expert Mode.”

What It Actually Costs You

On a representative 30K input / 5K output task, at cache-miss rates:

Model	Cost per task
DeepSeek V4-Flash (off-peak)	$0.0099
GPT-5.6 Luna	$0.0120
DeepSeek V4-Flash (peak)	$0.0198
DeepSeek V4-Pro (off-peak)	$0.0297
DeepSeek V4-Pro (old flat rate)	$0.0174
Gemini 3.7 Flash	$0.0410
DeepSeek V4-Pro (peak)	$0.0594
GLM-5.3	$0.0640
Grok 4.6	$0.0900

Off-peak V4-Flash remains the cheapest line on the board. Peak V4-Flash is 65% more expensive than Luna. Same model, same task, different hour.

The Timezone Arbitrage

Peak hours 01:00-04:00 and 06:00-10:00 UTC map to Chinese business hours. Translated:

US Eastern (UTC-4): peak is 21:00-00:00 and 02:00-06:00 — the entire American workday is off-peak.
US Pacific (UTC-7): peak is 18:00-21:00 and 23:00-03:00 — same story.
Central Europe (UTC+2): peak is 03:00-06:00 and 08:00-12:00 — mornings are hit, afternoons are free.

The schedule hits DeepSeek’s home market hardest and its export market lightest. If you are in North America and running interactive workloads, you may never touch a peak rate. If you are in Europe running batch jobs at 09:00 local, move them.

Why Now

The stated reason is resource allocation — encouraging users to “schedule their tasks based on actual usage.” The underlying reason is capacity. Demand is outrunning available compute, and DeepSeek is not the first to respond with price: Anthropic raised prices in April 2026 for the same reason.

The structural read is that DeepSeek has completed a two-tier product: Flash serves volume, Pro is priced for complexity. Peak/off-peak turns the clock itself into a pricing lever — which is a tool a supplier only reaches for when it has leverage.

What To Do

Check your job clock. 17 of 24 hours are half price. Anything deferrable — evals, batch summarization, index builds — should run off-peak. This is the single highest-leverage change.
Re-run your cache math. If your architecture was built around DeepSeek’s near-free cache hits, that assumption is now 6-12x more expensive. See prompt caching explained.
Re-benchmark against Luna and Gemini 3.7 Flash. At peak, the gap has inverted. See the post-hike budget comparison.
Don’t panic-migrate. Off-peak DeepSeek is still the cheapest option available, and interface compatibility (OpenAI + Anthropic formats) means switching later stays cheap.

Last verified: August 17, 2026, against DeepSeek’s official pricing documentation.