Flash example from Aug 17
For 1,000,000 cache-miss input tokens and 250,000 output tokens, the direct API estimate is $0.385 off-peak or $0.77 at peak. Retries, tool calls, taxes, and another provider's fees are outside that example.
One-month current-price offer
Lock in 20M tokens for $29 before prices rise
Current one-time pack prices end Sep 16, 11:59 p.m. Eastern Time. Pro rises from $29 to $35 on Sep 17 ET; the 20M token amount stays the same.
$29
Pro price through Sep 16 ET
$35
Pro price from Sep 17 ET
20M
same Pro token amount
One-time
no subscription or recurring charge
DeepSeek-V4.io is an independent service. The offer above applies to this site's one-time usage packs; DeepSeek Platform direct API pricing is documented separately below.
How billing works
Free boundary
A newly created account starts with 10,000 tokens. Sending stops when the balance reaches zero.
Paid boundary
A successful one-time checkout adds the selected number of tokens to the existing balance. There is no recurring charge.
Usage measurement
The service records provider-reported input and output token use. Your billing page shows balance changes and usage activity.
Three prices that should not be mixed
This website’s token packs pay for usage inside DeepSeek-V4.io. They are one-time purchases shown above, not an API key, a subscription, or a balance on DeepSeek Platform.
DeepSeek Platform API pricing is metered by the publisher from the input and output tokens processed by a specific API model. Developers obtain and fund that account directly with DeepSeek.
Third-party provider pricing belongs to that provider. Routing a DeepSeek model through another marketplace can introduce different rates, limits, caching rules, or fees, so its invoice cannot be calculated from DeepSeek’s direct table alone.
DeepSeek API price and cost reference
DeepSeek's official table was checked August 13, 2026. At 16:00 UTC on August 16—00:00 China Standard Time on August 17—the direct API moves to peak/off-peak billing. Off-peak rates are half the peak rates. Peak hours are 01:00–04:00 and 06:00–10:00 UTC. These are DeepSeek Platform prices, not a quote for this independent website.
| V4 Flash / 1M tokens | Through Aug 16 | Off-peak from Aug 17 | Peak from Aug 17 |
|---|---|---|---|
| Cache-hit input | $0.0028 | $0.007 | $0.014 |
| Cache-miss input | $0.14 | $0.22 | $0.44 |
| Output | $0.28 | $0.66 | $1.32 |
| V4 Pro / 1M tokens | Through Aug 16 | Off-peak from Aug 17 | Peak from Aug 17 |
|---|---|---|---|
| Cache-hit input | $0.003625 | $0.022 | $0.044 |
| Cache-miss input | $0.435 | $0.66 | $1.32 |
| Output | $0.87 | $1.98 | $3.96 |
Source: DeepSeek Models & Pricing and the August 13 change log. Check the live table before budgeting or topping up.
How to estimate direct API cost
Calculate cache-hit input, cache-miss input, and output separately, then add the three amounts. For each category: tokens ÷ 1,000,000 × rate. Do not price every input token at the cache-hit rate unless the provider reports it as a hit.
Pro example from Aug 17
For the same token counts, the direct API estimate is $1.155 off-peak or $2.31 at peak. This illustrates the formula, not a quote for this website or a guarantee that both models produce identical output lengths.
DeepSeek V4 Pro price notes that change a real bill
- Output often dominates. Long reasoning traces, generated code, and verbose answers can make output the largest line item even when the prompt is short.
- A 1M context limit is not a target. Sending more history increases tokens, latency, and the chance that important instructions are diluted. Trim or retrieve only the evidence needed for the task.
- Caching is conditional. Stable repeated prefixes may qualify for a lower input rate, but changing system prompts or document order can prevent a hit. Use provider usage fields rather than assuming savings.
- Thinking mode changes usage. Compare answer quality and total reported tokens on your own workload. A lower per-token model can still cost more if it needs more retries or longer outputs.
- Concurrency is not token volume. The official snapshot lists concurrency limits of 2,500 for Flash and 500 for Pro; it does not mean a request is free, instantaneous, or guaranteed admission.
DeepSeek free API key, trial, and paid boundaries
This page does not advertise a free DeepSeek API key. A DeepSeek Platform key belongs to the platform account that created it, and any granted balance or promotion is governed by the publisher’s current account terms. DeepSeek’s deduction rules say granted balance is preferred when both granted and topped-up balances exist, but the presence or amount of a grant is not promised here.
The 10,000-token signup balance described on this page is different: it is an application balance for trying DeepSeek-V4.io. It cannot be exported, exchanged for cash, or used as credentials against DeepSeek’s direct API. When it reaches zero, the next step is either a one-time pack for this application or a separately funded provider API account, depending on whether you need the website workflow or your own integration.
A practical budgeting checklist
- Choose the exact product first: this website, DeepSeek Platform, or a third-party provider.
- Record the exact model identifier, mode, price-table date, and currency.
- Measure input, cache-hit input, output, retries, and failed calls from real usage logs.
- Estimate a low, expected, and high monthly case instead of multiplying one demo request.
- Include engineering, storage, retrieval, monitoring, and human-review costs when comparing alternatives.
- Set balance and usage alerts, then revisit the official price page before each major rollout.
Pricing FAQ
Buy before the current prices end
Get 20M Pro tokens for $29 through Sep 16, 11:59 p.m. Eastern Time. The same pack becomes $35 on Sep 17 ET.
20M Pro tokens for $29
$35 from Sep 17 ET
One-time purchase · no subscription