API pricing · 9 min read

Reprice the DeepSeek V4 event on the correct UTC and cache basis

DeepSeek’s V4 peak/off-peak schedule is announced to start August 16, 2026 at 16:00 UTC. The cheapest window is not the default: schedule flexible batch work outside the 01:00–04:00 and 06:00–10:00 UTC peak windows, and label every estimate by rate basis until activation.

By Todd · Published 2026-08-14 · Verified 2026-09-02

DeepSeekOff-peakCost controlScheduling

Effective	2026-08-16 at 16:00 UTC
Peak UTC	01:00–04:00 and 06:00–10:00
V4 Flash off-peak	$0.007 / $0.22 / $0.66 per M
V4 Pro off-peak	$0.022 / $0.66 / $1.98 per M
Rate basis	Cache hit / cache miss / output
Verified	2026-09-02

Effective-date banner

These are announced rates, effective August 16 at 16:00 UTC. Before that timestamp, do not describe the new schedule as current. After activation, resolve the rate from the event time rather than from deploy time.

Peak and off-peak windows

DeepSeek identifies 01:00–04:00 UTC and 06:00–10:00 UTC as peak windows. All other UTC hours are off-peak. The boundary matters: 04:00 UTC is off-peak, 06:00 UTC is peak, and 10:00 UTC is off-peak.

V4 Flash peak: $0.014 cache-hit, $0.44 cache-miss input, $1.32 output per million
V4 Flash off-peak: $0.007 cache-hit, $0.22 cache-miss input, $0.66 output per million
V4 Pro peak: $0.044 cache-hit, $1.32 cache-miss input, $3.96 output per million
V4 Pro off-peak: $0.022 cache-hit, $0.66 cache-miss input, $1.98 output per million

Scheduling guidance

Put flexible batch jobs, embedding-adjacent enrichment, nightly evaluations, and large cacheable prompts into off-peak UTC hours. Do not trade away an SLA to save tokens: latency, concurrency, provider throttling, retries, and queue time can dominate the nominal rate difference. Convert UTC to the queue’s actual timezone and store the UTC schedule as the source of truth.

Cost examples

At off-peak rates, 10M cache-hit input tokens plus 1M output tokens cost about $0.73 on V4 Flash. The same workload with 10M cache-miss input tokens costs about $2.86. On V4 Pro, the corresponding totals are about $2.20 and $8.58. These examples exclude retries and any provider-specific minimums.

Caveats

A cache hit is not the same thing as a cheaper cache miss, and an off-peak rate is not a promise of lower latency. Record cache status, UTC event time, and retry count with every cost sample so your finance dashboard does not blend incomparable rates.

Explore the canonical data

DeepSeek V4 Flash model page →DeepSeek V4 Pro model page →DeepSeek provider pricing →V4 Flash pricing →V4 Pro pricing →Flash vs Pro comparison →LLM cost calculator →

Test it yourself

Run the same prompt across deepseek v4 flash, deepseek v4 pro in one workspace.

Open the comparison playground →

FAQ

When does DeepSeek V4 off-peak pricing start?

The announced schedule becomes effective August 16, 2026 at 16:00 UTC.

Which hours are peak?

01:00–04:00 UTC and 06:00–10:00 UTC. Other hours are off-peak.

Sources

DeepSeek V4 peak/off-peak event arithmetic

The DeepSeek V4 pricing article preserves the dated rate change, UTC peak boundaries, cache token classes, and frozen before/after arithmetic. It does not forecast arbitrary workloads or latency. Verified 2026-09-02; missing event time, cache status, retry evidence, or rate version makes repricing Unavailable.

Verification identity: Luna / Continuous SEO Builder Batch 57 · verifiedAt 2026-09-02 · exact route owner deepseek-v4-off-peak-pricing. Qualitative demand is search-result evidence only; exact monthly volume is unavailable.

Price-event source and effective-time register

Scope: Owns announcement, pre-effective, activation, UTC peak boundary, and later-document events without applying a later table retroactively.

Deterministic formula/rule: applicableRate = latest source with effectiveAt <= eventAt and matching model/rate version; otherwise Unavailable

Evidence owner: DeepSeek pricing documentation · verified 2026-09-02. Missing or conflicting joins fail closed.

Frozen scenario	Request / response / model / provider / host / surface / realm / account / artifact / claim / rate / event / prompt / config / usage fields	Bounded result
`batch57-deepseek-v4-off-peak-pricing-m1-r1` announcement	Inputs: one dated event or source artifact; fixture=announcement Joined fields: event ID, artifact URL/hash, publisher, exact endpoint or model ID, event/capture/verification timestamps, current/superseded/missing state; scenario-specific evidence key=announcement Verification: observed values are accepted only from the named source; assumptions remain labeled.	DATED — artifact/date join required Keep the event snapshot separate from current state and use the narrowest source-backed date Fixture outcome is bounded to announcement; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m1-r2` pre-effective	Inputs: one dated event or source artifact; fixture=pre-effective Joined fields: event ID, artifact URL/hash, publisher, exact endpoint or model ID, event/capture/verification timestamps, current/superseded/missing state; scenario-specific evidence key=pre-effective Verification: observed values are accepted only from the named source; assumptions remain labeled.	DATED — artifact/date join required Keep the event snapshot separate from current state and use the narrowest source-backed date Fixture outcome is bounded to pre-effective; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m1-r3` exact activation	Inputs: one dated event or source artifact; fixture=exact activation Joined fields: event ID, artifact URL/hash, publisher, exact endpoint or model ID, event/capture/verification timestamps, current/superseded/missing state; scenario-specific evidence key=exact activation Verification: observed values are accepted only from the named source; assumptions remain labeled.	DATED — artifact/date join required Keep the event snapshot separate from current state and use the narrowest source-backed date Fixture outcome is bounded to exact activation; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m1-r4` peak start	Inputs: one dated event or source artifact; fixture=peak start Joined fields: event ID, artifact URL/hash, publisher, exact endpoint or model ID, event/capture/verification timestamps, current/superseded/missing state; scenario-specific evidence key=peak start Verification: observed values are accepted only from the named source; assumptions remain labeled.	DATED — artifact/date join required Keep the event snapshot separate from current state and use the narrowest source-backed date Fixture outcome is bounded to peak start; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m1-r5` peak end	Inputs: one dated event or source artifact; fixture=peak end Joined fields: event ID, artifact URL/hash, publisher, exact endpoint or model ID, event/capture/verification timestamps, current/superseded/missing state; scenario-specific evidence key=peak end Verification: observed values are accepted only from the named source; assumptions remain labeled.	DATED — artifact/date join required Keep the event snapshot separate from current state and use the narrowest source-backed date Fixture outcome is bounded to peak end; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m1-r6` later document change	Inputs: one dated event or source artifact; fixture=later document change Joined fields: event ID, artifact URL/hash, publisher, exact endpoint or model ID, event/capture/verification timestamps, current/superseded/missing state; scenario-specific evidence key=later document change Verification: observed values are accepted only from the named source; assumptions remain labeled.	DATED — artifact/date join required Keep the event snapshot separate from current state and use the narrowest source-backed date Fixture outcome is bounded to later document change; no broader claim is inferred.

Peak/off-peak cache-basis reconciliation grid

Scope: Owns Flash/Pro cache-hit, cache-miss, and output cells with compatible rate windows and token units; blended cells fail closed.

Deterministic formula/rule: cost = tokens / 1,000,000 × documentedRate(tokenClass, model, window); missing class/time => ambiguous

Evidence owner: DeepSeek pricing documentation · verified 2026-09-02. Missing or conflicting joins fail closed.

Frozen scenario	Request / response / model / provider / host / surface / realm / account / artifact / claim / rate / event / prompt / config / usage fields	Bounded result
`batch57-deepseek-v4-off-peak-pricing-m2-r1` Flash cache hit	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=Flash cache hit Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=Flash cache hit Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to Flash cache hit; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m2-r2` Flash cache miss	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=Flash cache miss Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=Flash cache miss Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to Flash cache miss; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m2-r3` Flash output	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=Flash output Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=Flash output Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to Flash output; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m2-r4` Pro cache hit	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=Pro cache hit Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=Pro cache hit Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to Pro cache hit; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m2-r5` Pro cache miss	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=Pro cache miss Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=Pro cache miss Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to Pro cache miss; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m2-r6` Pro output	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=Pro output Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=Pro output Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to Pro output; no broader claim is inferred.

Frozen-workload repricing and scheduling board

Scope: Owns before/after arithmetic for a frozen workload and schedule decision; price alone cannot establish latency, reliability, or SLA outcome.

Deterministic formula/rule: delta = oldTotal - newTotal; outcome = schedule only when time/tokens/retry assumptions are labeled

Evidence owner: DeepSeek pricing documentation · verified 2026-09-02. Missing or conflicting joins fail closed.

Frozen scenario	Request / response / model / provider / host / surface / realm / account / artifact / claim / rate / event / prompt / config / usage fields	Bounded result
`batch57-deepseek-v4-off-peak-pricing-m3-r1` fully flexible batch	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=fully flexible batch Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=fully flexible batch Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to fully flexible batch; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m3-r2` SLA-bound traffic	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=SLA-bound traffic Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=SLA-bound traffic Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to SLA-bound traffic; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m3-r3` mixed cache status	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=mixed cache status Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=mixed cache status Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to mixed cache status; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m3-r4` boundary-spanning job	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=boundary-spanning job Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=boundary-spanning job Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to boundary-spanning job; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m3-r5` retry-heavy job	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=retry-heavy job Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=retry-heavy job Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to retry-heavy job; no broader claim is inferred.
`batch57-deepseek-v4-off-peak-pricing-m3-r6` missing event time	Inputs: one frozen workload, billing, or endpoint-resolution fixture; fixture=missing event time Joined fields: request/log/invoice or input/output hashes, exact model/provider/host, token class, event time/timezone, rate version, retries, observed-versus-assumed marker; scenario-specific evidence key=missing event time Verification: observed values are accepted only from the named source; assumptions remain labeled.	FAIL CLOSED — exact workload join required Calculate or schedule only from the joined observed fields; any missing identity, time, rate, tokenizer, cache, or invoice field returns Unavailable Fixture outcome is bounded to missing event time; no broader claim is inferred.

Owner-attributed next step: run a matched comparison at /try?source=batch57-deepseek-v4-off-peak-pricing. Historical and unavailable states remain visible until their exact source joins resolve.