About the DeepSeek Pricing Calculator

This calculator is for developers, product teams, and FinOps practitioners who need an inspectable estimate before sending traffic to the DeepSeek API. It shows which pricing period applies, what a request may cost, how V4 Flash compares with V4 Pro, and what delay-tolerant work might save off-peak.

There is no account, upload, or connection to a DeepSeek workspace. You enter a request profile in the browser and receive a deterministic estimate from published rates. Cache-hit input, cache-miss input, and output remain separate because they are billed differently. Time also stays visible so a forecast does not quietly assign traffic to the wrong period.

The goal is planning, not a promise about an invoice. Use the result in a design review, compare schedules, or sanity-check a budget. Once a service is live, replace assumptions with observed usage and official billing records.

A flat rate lets a team forecast from tokens alone; peak pricing adds time. Otherwise identical requests can cost different amounts when one falls inside a peak window. That matters for agent loops, large-context analysis, evaluations, and output-heavy pipelines.

DeepSeek's new schedule sets off-peak rates at half of peak rates. Real-time actions can keep their latency target, while batch summaries, tests, indexing, reports, and offline evaluation may enter a deadline-based queue. The useful question is not “Can everything wait?” but “What share can move without degrading the product?” Monthly mode models that share instead of assuming an all-or-nothing migration.

Rates, model details, effective time, and peak windows come from DeepSeek's official API documentation. At the last review, it listed deepseek-v4-flash as DeepSeek-V4-Flash-0731 and deepseek-v4-pro as DeepSeek-V4-Pro-0813. New pricing begins at 16:00 UTC on August 16, 2026. Peak periods are 01:00–04:00 and 06:00–10:00 UTC; all remaining hours are off-peak.

From that effective time, Flash off-peak rates per 1 million tokens are $0.007 for cache-hit input, $0.22 for cache-miss input, and $0.66 for output; its peak rates are $0.014, $0.44, and $1.32. Pro off-peak rates are $0.022, $0.66, and $1.98; its peak rates are $0.044, $1.32, and $3.96.

The change log says the older deepseek-chat and deepseek-reasoner API names were scheduled for discontinuation on July 24, 2026. They are not separate choices in this calculator. The current scope follows the two V4 identifiers in the active official price table instead of preserving historical aliases that could mislead someone planning new traffic.

The official change log records the adjustment and provides release context. Prices can change, so the application uses a reviewed configuration instead of scraping documentation during your visit. An external layout edit should never become an unreviewed billing calculation.

The calculator divides input into cache hits and misses using your chosen ratio. Each group is divided by 1 million and multiplied by its applicable rate; output is calculated separately. Adding the three amounts gives cost per request, which is multiplied by request count in single-request mode.

Monthly requests equal requests per day times billing days. Peak share divides that volume between peak and off-peak request costs, and the two contributions produce the estimate. Potential savings subtract an all-off-peak scenario from that mixed total. For traffic spread evenly across a day, the default peak share is 7/24, or about 29.2%.

Changing one field at a time makes scenario results easier to interpret. Hold tokens and traffic constant when comparing models, hold the model constant when testing a schedule, and vary cache ratio only when you have a realistic reason to expect different prefix reuse. This keeps unrelated assumptions from hiding the factor you intended to measure.

The calculation keeps more precision than the interface normally shows. Small request values receive extra decimal places so a real cost does not appear as $0.00, and totals use unrounded values.

Classification always starts in UTC. Lower boundaries are included and upper boundaries excluded: 01:00:00 is peak, 04:00:00 is off-peak, 06:00:00 is peak, and 10:00:00 is off-peak.

The browser timezone uses an IANA name such as America/New_York or Asia/Singapore. IANA rules apply the correct date-specific daylight saving offset without changing DeepSeek's UTC windows. If detection fails, the tool uses UTC and presents a notice instead of guessing. Timezone changes the schedule display, not the US-dollar currency.

The scope is official DeepSeek API pricing for deepseek-v4-flash and deepseek-v4-pro. Single and monthly estimates include cache-hit input, cache-miss input, output, model, volume, and the applicable peak or off-peak period. A single request can use now or a custom time; monthly mode accepts an even or user-supplied peak share.

You can compare Flash with Pro, see the peak premium, estimate a month, and model off-peak savings. Comparisons hold token usage constant to isolate price. They do not claim the models will produce the same output length or result.

The tool does not price OpenRouter, cloud platforms, gateways, or resellers. They can set their own rates, currencies, discounts, taxes, and timing rules. It also excludes self-hosted GPU, storage, networking, staff, observability, and queueing costs. It is not a proxy and does not sell credits or submit requests.

Results stay in US dollars and exclude tax. The calculator does not inspect prose, infer tokens from a prompt, predict retries, or forecast output behavior. It provides no financial, legal, or investment advice. It never needs an API key, prompt, model response, account identifier, or billing statement.

The formula calculates the supplied scenario deterministically, but that scenario may not match production. Error often comes from average tokens, cache assumptions, timing, retries, growth, or a different provider. Compare actual usage with your inputs; mature services may need separate profiles for short and long requests.

Pricing is checked manually against official documentation. A change requires configuration review, automated arithmetic and boundary tests, and redeployment. “Last verified” records the check date, not a guarantee of no later change. Review the official links before a material commitment; DeepSeek may adjust prices, and its invoice remains authoritative.

Availability, latency, concurrency, tokenization, and output quality are separate from price. Use the estimate alongside service goals, provider limits, and operational evidence.

In production, Google Analytics 4 measures visits and broad calculator interactions, while Microsoft Clarity helps assess usability. When configured, they load with the page without a custom consent prompt. Browser, platform, regional, or legal controls may still apply. See Google and Microsoft privacy information for their processing details.

Calculator fields are masked in Clarity replay. Events use broad properties such as model, mode, pricing period, or ranges. We do not intentionally send exact token entries, prompts, API keys, account data, or complete calculation URLs. Google Search Console and Bing Webmaster Tools measure search performance; verification tags establish site ownership without connecting a DeepSeek account.

Non-sensitive settings may remain in your browser so a refresh does not erase useful work. That local state is not uploaded to build a profile. Never enter secrets: the calculator needs counts and scheduling assumptions, not prompts or credentials.