Home › Blog › DeepSeek split into two models with peak-hour pricing

DeepSeek Split Into Two Models With Peak-Hour Pricing — We Still Had the Old Flat Rate

12 November 2026 · Site Methodology & Corrections · 5 min read

I went to our DeepSeek page to add a cheaper-alternatives link and noticed the pricing table still said "DeepSeek-V3 (chat)" and "DeepSeek-R1 (reasoner)" at a flat $0.27/$1.10 and $0.55/$2.19 per 1M tokens — numbers dated July 2026. DeepSeek's own current pricing page doesn't mention either name anymore. It lists exactly two models, with a billing dimension our calculator doesn't even have a field for: time of day.

What DeepSeek's own docs say today

api-docs.deepseek.com now lists deepseek-flash (served by DeepSeek-V4.1-Flash) and deepseek-v4-pro (DeepSeek-V4-Pro) — nothing else. Both are billed on a peak/off-peak schedule: off-peak is the baseline rate, and peak hours cost exactly double. Peak is defined as 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday, excluding Chinese public holidays — so every weekend, every holiday, and most of each weekday is actually billed at the lower rate.

ModeloInput (off-peak / peak)Cache-hit inputOutput (off-peak / peak)
DeepSeek-V4.1-Flash$0.15 / $0.30$0.003$0.60 / $1.20
DeepSeek-V4-Pro$0.66 / $1.32$0.022$1.98 / $3.96

Legacy model-ID strings like deepseek-v4-flash e deepseek-v4-flash-vision-exp still work — the docs say they're "served by DeepSeek-V4.1-Flash-1017 and billed at the Flash price" — but there's no mention anywhere of deepseek-chat ou deepseek-reasoner, the model IDs tied to the old V3/R1 names. Three independent sources (DeepSeek's own pricing and model-list pages, plus third-party trackers at mercatus-ai.com and aipricing.guru) agree on the $0.15/$0.60 and $0.66/$1.98 off-peak numbers, so this isn't a one-page fluke.

O que isso faz com uma conta real

Take a workload of 50M input and 10M output tokens a month — a mid-size chatbot. Here's what it would have cost on our old stale number versus what DeepSeek actually charges now, off-peak and at peak:

CenárioRate usedCusto mensal
Our old number (stale "V3")$0.27 / $1.10$24.50
Flash, off-peak$0.15 / $0.60$13.50
Flash, peak$0.30 / $1.20$27.00
V4-Pro, off-peak$0.66 / $1.98$52.80
V4-Pro, peak$1.32 / $3.96$105.60

So depending on which current model you actually land on and when your traffic runs, the real number is anywhere from 45% cheaper to 4.3x more expensive than what our page was quoting. Flash off-peak undercuts our stale figure comfortably — DeepSeek got cheaper for anyone who avoids the two daily UTC windows. But anyone reasoning our old "DeepSeek-R1" line mapped onto what's now V4-Pro was looking at a number 2.4x too low off-peak, and over 4x too low at peak.

What we changed

Updated the DeepSeek pricing page to the current two-model line-up with the off-peak baseline and an explicit peak-hour warning, refreshed the shared models.js entries that feed our calculadora de custos and the individual DeepSeek V4 (Pro) e DeepSeek V4 Flash comparison pages across all 7 hosts we track, and set "last verified" to today. We kept the off-peak rate as the default number rather than peak, because it covers more actual hours of the week — but we say so explicitly next to the price instead of leaving it implicit like the old page did.

We didn't touch the open-weight reseller rows (Together, Fireworks, DeepInfra, Groq, Novita, OpenRouter) for the retired V3/R1/Coder names — those hosts serve the open model weights independently of whatever DeepSeek's own direct API calls them this month, so their pricing isn't tied to this rename.

Why this is easy to miss

A per-token price doesn't carry a timestamp. "$0.27 input, $1.10 output" reads exactly the same whether DeepSeek set that number yesterday or four months ago, and nothing about a static HTML table signals that the vendor quietly restructured its entire pricing model underneath it — new names, a new tier, and a time-of-day dimension that a flat-rate calculator has no field for. The only way to catch it is to go re-check the source instead of trusting what's on file, which is exactly what this routine audit pass is for.

Estimate your own DeepSeek workload on the Calculadora de custos de API de IA, compare it against OpenAI, Anthropic and Google on the model price comparison page, or see the full write-up on DeepSeek API pricing. New to usage-based AI pricing? Start with the guias gratuitos de custo de API.

Pricing sourced from DeepSeek's own API documentation (api-docs.deepseek.com) plus two independent third-party price trackers, checked 12 November 2026. Off-peak figures are the baseline we display; confirm current peak-hour rates on the provider's own page before budgeting continuous-traffic workloads.