DeepSeek V4.1 Flash API pricing: cache, peak hours and cost

DeepSeek V4.1 Flash has three token prices, not one: cached input, uncached input and output. The official API also applies different rates by time of day. For a useful budget, identify the model, billing provider and workload before multiplying tokens by a headline price.

Rates below were checked against DeepSeek’s official pricing page on September 10, 2026. They are USD prices for direct DeepSeek API use, not Ofox or another gateway’s quote. No paid request or billing-recovery test was performed for this article.

Official rates per one million tokens

The listed model is deepseek-flash, currently V4.1 Flash. If you are still sending a V4-era name, check the release and alias summary before interpreting an old price table. Pro’s September 14 change has a separate migration checklist.

Do not apply the cache-hit price to all input. A repeated instruction is not, by itself, evidence that your entire request was billed as a cache hit. Read the returned usage fields and the provider’s billing record.

When do peak prices apply?

Official peak periods are Monday through Friday, 01:00–04:00 and 06:00–10:00 UTC. Other periods are off-peak. The same periods are 09:00–12:00 and 14:00–18:00 in Beijing, and 10:00–13:00 and 15:00–19:00 in Tokyo and Seoul.

Use a timezone-aware conversion for a location that changes clocks seasonally. Do not label one fixed local schedule as correct worldwide. If a request crosses a pricing boundary, consult the provider’s billing rule and actual record rather than inventing a split-charge formula.

A request budget you can change

Consider a deliberately simple workload: 100 requests, each with 10,000 uncached input tokens and 2,000 billed output tokens. That totals 1 million uncached input tokens and 0.2 million output tokens.

Off-peak: (1 × $0.15) + (0.2 × $0.60) = $0.27
Peak:     (1 × $0.30) + (0.2 × $1.20) = $0.54

This is arithmetic under explicit assumptions, not a measured average agent cost. If an agent makes several model calls, retries, accumulates conversation history or generates more billed output, replace the assumed totals with those actual quantities. Do not estimate from the final answer’s visible length alone.

For a mixed-cache workload, use:

Cost = cached_input_tokens / 1_000_000 × cached_input_rate
     + uncached_input_tokens / 1_000_000 × uncached_input_rate
     + billed_output_tokens / 1_000_000 × output_rate

The same example would fit about 18 complete 100-request batches into a $5 off-peak token budget, or 9 at peak rates. That excludes any provider-specific fees and assumes all 100 requests meet the stated token budget. It is not a promise that a $5 purchase always buys 1,800 useful tasks.

Compare providers on a whole task

Before purchasing credit, record the actual model ID, supported operation, displayed rates and payment conditions. A provider can use a different identifier, billing unit or routing policy from the official endpoint. Converting currency yourself does not create a local-currency vendor quote.

For coding agents, compare the cost of finishing the same small task, including corrective turns. The Codex setup and Claude Code setup explain how configuration affects which model receives the work. A cheaper selected main model does not establish what every helper or subagent uses.

If you are evaluating Ofox, check its catalog and route price, then create an account if the required model and operation are supported. Start from a small task budget appropriate to that provider. Registering or buying credit elsewhere does not add balance to your direct DeepSeek account.

What to verify after the first call

Use the API setup guide for a minimal request. Save sanitized usage totals, model, provider, time and billed amount. Compare that record to the formula before scheduling a batch. Where a discrepancy remains, identify whether it comes from cache classification, output usage, call count, time band or the provider’s own pricing rather than assuming the arithmetic table is a billing guarantee.

Before choosing where to fund access, use the provider buying checklist to confirm the model, protocol and billing terms.