DeepSeek V4.1 Flash pricing: peak/off-peak, cache-hit rate, how the bill adds up
Peak, cache hit and how the bill adds up
As of 2026-10-19 the live DeepSeek version is DeepSeek-V4.1-Flash-1017, priced per 1M tokens with off-peak at half of peak; 1M context and 384K max output. All six rates and the peak windows, converted to your local timezone, are below.
Updated 2026-10-20
Prices change: on QCode, Claude, GPT and Chinese models are billed per token — live rates on /models.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Six rates and two limits
Input (cache miss)
$0.15 off-peak, $0.3 peak, per 1M tokens; the live version is DeepSeek-V4.1-Flash-1017.
Output
$0.6 off-peak, $1.2 peak, per 1M tokens.
Input (cache hit)
Cache-hit input drops to $0.003 off-peak / $0.006 peak.
Concurrency
Flash 2500; V4 Pro on the same page is 500.
What this is
As of 2026-10-19 the official sheet quotes per 1M tokens and the live version is DeepSeek-V4.1-Flash-1017: input splits into cache-hit and cache-miss rates, then into off-peak and peak. Off-peak is half of peak; peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday (excluding Chinese public holidays), everything else off-peak.
What happened
On 2026-10-17 DeepSeek released DeepSeek-V4.1-Flash-1017; the same-day changelog states that the API calling method and pricing remain unchanged — simply use deepseek-flash to access the latest model — and the official web and app models moved to the new version too. DeepSeek calls the release a Pareto improvement in performance and cost: stronger benchmark performance, faster task completion and shorter outputs than the previous version. New pricing took effect 04:00 UTC on 2026-09-10, when V4.1 Flash shipped; the changelog's 2026-09-10 entry confirmed V4 Pro continues with unchanged billing.
Timeline
2026-08-13 DeepSeek-V4-Pro-0813 went GA — the V4 Pro column on today's sheet.
2026-09-10 04:00 UTC V4.1 Flash live with the new pricing (from $0.15 / $0.6).
The changelog's 2026-09-10 entry confirms V4 Pro continues, billing unchanged. On 2026-10-17 the official changelog said the API calling method and pricing remain unchanged — use deepseek-flash to reach the latest model.
Confirmed vs caution
Officially confirmed
The pricing page names the live versions DeepSeek-V4.1-Flash-1017 and DeepSeek-V4-Pro-0813 (GA 2026-08-13), and states that off-peak is half of peak, with peak hours 01:00-04:00 and 06:00-10:00 UTC on weekdays (excluding Chinese public holidays).
⚠️ Caution
No end date is published; the footnote says prices may change and the page is authoritative. These three rates are the ones in force since 2026-09-10 04:00 UTC — don't lock them into a long-term customer quote. DeepSeek's pricing page states that on its own API the legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp map to retired models, with requests served by DeepSeek-V4.1-Flash-1017 and billed at the Flash price (that is DeepSeek's API; QCode's model list follows /models).
Same vs differs
Same
Both priced per 1M tokens, three rates (cache-hit input, cache-miss input, output), both split peak/off-peak, off-peak half of peak.
Differs
Cache-miss input $0.66 (off-peak) against $0.15 at Flash; output $1.98 against $0.6; concurrency 500 against 2500; Flash takes vision input, V4 Pro does not.
What to do
Apply the three rates to your own input/output mix, then decide whether to shape long context as a cache-hittable prefix; repeating the same prompt drops input cost by an order of magnitude. Since 2026-10-17 the live version is DeepSeek-V4.1-Flash-1017, with no change to how it is called or what it costs.
On QCode
On QCode set deepseek-v4.1-flash — billing is per token, and the per-model prices are listed on /models. To compare, switch the same key to deepseek-v4-pro (from $0.66 / $1.98). Not ready for a monthly plan? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days.
FAQ
What are the six rates?
Cache-miss input $0.15 off-peak / $0.3 peak; output $0.6 / $1.2; cache-hit input $0.003 / $0.006 — all per 1M tokens. The live version is DeepSeek-V4.1-Flash-1017.
When is peak?
Peak is 01:00-04:00 and 06:00-10:00 UTC, Monday through Friday, excluding Chinese public holidays; all other hours are off-peak, weekends included. That is 09:00-12:00 and 14:00-18:00 Beijing, 10:00-13:00 and 15:00-19:00 Tokyo, 04:00-07:00 and 09:00-13:00 Moscow.
How much cheaper than V4 Pro?
Cache-miss input $0.15 against $0.66, output $0.6 against $1.98, cache-hit input $0.003 against $0.022. All off-peak; at peak V4 Pro is $1.32 for cache-miss input, $3.96 for output and $0.044 for cache-hit input. Flash allows 2500 concurrent requests against 500, and Flash takes vision input while V4 Pro does not.
When does cache hit pay off?
A hit takes input from $0.15 to $0.003 — 1/50. Put the unchanging system prompt and long documents at the front and vary only the tail.
Does thinking mode cost more?
Thinking emits extra thinking tokens billed as output. At the same $1.2/M peak output rate, a longer chain of thought means a bigger bill — set a non-thinking baseline first.
Will this price hold?
The footnote reserves the right to adjust and points back to the official page, which is why this page carries a check date of 2026-10-19; the official changelog says the 2026-10-17 upgrade left both the calling method and the prices unchanged. Re-check the official pricing page before quoting long term.
Sources
DeepSeek official Models & Pricing, changelog and the 2026-10-17 release post, captured 2026-10-19. All rates per 1M tokens.
Estimate your own bill on per-token billing
One QCode key: set deepseek-v4.1-flash (availability and version follow /models), and switch to deepseek-v4-pro to compare.
Related
Claude Code auto mode guide
Interactive sessions now default to auto mode (docs say 2.1.283, changelog 2.1.284): the six permission modes, how to turn it off, and notes for teams and CI.
V4.1 Flash vs V4 Pro
Which tier to pick.
Claude Code 1M context behind a custom base URL
From 2.1.285, a custom base URL gets 1M automatically; use /autocompact 200k if your gateway stops at 200K, plus hint headers and 529 backoff.
Crypto payment guide
Nine coins, 56 networks, zero platform fee for AI coding API credits.
Checked on 2026-10-19. Only the official rate table and the time windows are restated — no third-party benchmarks, no speculation. DeepSeek reserves the right to adjust pricing, and the official pricing page is authoritative. Which models you can call is listed on /models.