DeepSeek V4.1 Flash Is Half Price Off-Peak, but Most Hosts Charge the Peak Rate

DeepSeek charges half price for DeepSeek V4.1 Flash outside its peak hours: $0.15 per million input tokens and $0.60 per million output tokens instead of $0.30 / $1.20. Those off-peak hours cover 79% of the week. Together, Novita and Fireworks list $0.30 / $1.20 around the clock. So if you use V4.1 Flash through one of them, you pay DeepSeek's peak rate at 3 a.m. too.

The off-peak rate is not in the open datasets we track, so our model page shows only the peak price. This take fills that gap.

The numbers

One realistic month for a product with an AI feature: 10 million input tokens and 2 million output tokens, no cache hits.

Where	Input / 1M	Output / 1M	Cost for that month
DeepInfra, Flex tier during its promotion	$0.112	$0.336	$1.79
DeepInfra, promotion (30% off)	$0.14	$0.42	$2.24
DeepSeek, off-peak	$0.15	$0.60	$2.70
DeepInfra, regular price	$0.20	$0.60	$3.20
DeepSeek, peak	$0.30	$1.20	$5.40
Together AI	$0.30	$1.20	$5.40
Novita AI	$0.30	$1.20	$5.40
Fireworks AI	$0.30	$1.20	$5.40

Prices from the official pages, checked on November 1, 2026. The cost column is our arithmetic. DeepInfra's Flex tier is 0.8x its standard price; the page does not say when the promotion ends.

In our tracking, six of the eight hosts that sell V4.1 Flash list exactly $0.30 / $1.20. Fireworks charges $0.45 / $1.80 if your requests must stay in the US.

When off-peak is

DeepSeek's peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. Everything else is off-peak, including whole weekends. That is 35 peak hours out of 168, so 133 hours or 79% of the week are half price.

Where you are decides how much of that you get:

• US East Coast: peak is 20:00 to 23:00 and 01:00 to 05:00 Eastern time. Your whole working day is off-peak.
• Central Europe: peak is 02:00 to 05:00 and 07:00 to 11:00 CET. Your morning traffic pays the peak rate, the afternoon does not.
• Batch jobs: schedule them outside the two windows and every token is half price.

Caching makes the gap smaller, not bigger

V4.1 Flash is cheap on cache hits everywhere: DeepSeek charges $0.003 off-peak and $0.006 at peak, DeepInfra $0.004 during its promotion, Together $0.006. If most of your input is a repeated system prompt, the output price decides the bill, and there DeepInfra ($0.42 promotional, $0.60 regular) and DeepSeek off-peak ($0.60) are well ahead of the $1.20 that Together, Novita and Fireworks charge.

What I would do

• You run V4.1 Flash on a host that lists $0.30 / $1.20: you pay the peak rate all day. Moving to DeepInfra or to DeepSeek's off-peak hours cuts that bill roughly in half.
• Your traffic is mostly US daytime or batch work: use DeepSeek directly. In practice you get $0.15 / $0.60 for most requests.
• Your traffic peaks in the European morning, or you cannot send data to DeepSeek's own API: DeepInfra is cheaper than DeepSeek's peak rate even at its regular price. Treat the promotion as a bonus, not as the price you plan with.

Before you switch, test your own prompts on the new endpoint. We compare list prices; we do not measure speed, rate limits or output quality. Price your own volume on the DeepSeek V4.1 Flash page, and see how we track prices.

The prices behind this take

Checked by hand on November 1, 2026: DeepSeek API pricing (peak and off-peak rates, peak hours; the legacy name deepseek-v4-flash is now served and billed as V4.1 Flash), DeepInfra's DeepSeek V4.1 Flash page (promotional, regular and Flex prices), Together AI pricing, Novita AI pricing and Fireworks AI serverless pricing.

The honest summary: the same model costs the same tokens everywhere, but not at the same hour. Know when your traffic runs, and do not pay the peak rate at night.