DeepSeek Token Counter
Estimate your text in DeepSeek tokens and price it on both V4 models. DeepSeek is the only provider on this site that bills by the clock — requests made during its off-peak windows cost half the listed rate — and it has the steepest cache discount in the whole catalogue, at $0.044 per 1M cached tokens against $1.32 standard.
- Countingest.
- Context1M
- Input$1.32
- Output$3.96
- Cached$0.0443.3% of input
Estimate DeepSeek tokens
Opens on DeepSeek V4 Pro. Both V4 rows share the same 1,000,000-token window, so switching between them changes only the price — by a factor of exactly three.
Drop files here or click to browse
Supports PDF, TXT, MD, JSON, CSV, XML — max 10MB
Count as an API request
Providers also bill the role and boundary tokens that wrap your text. Anthropic's own example — a short system prompt plus one message — costs 14 input tokens for 7 tokens of visible text.
How accurate is this DeepSeek token count?
It is an estimate. DeepSeek does not publish the tokenizer its V4 models use and offers no token-counting endpoint, so there is no authoritative number available before you send the request — the usage figures in the API response are the first exact count you get. The estimate here is calibrated on English prose and, as with every approximation of this kind, is least reliable on code, JSON and Chinese text, where DeepSeek is often used most.
The practical consequence is to leave headroom. If a workflow depends on staying under a size limit, size it against the estimate with a margin rather than to the token, and read the usage block that comes back with each response — it reports prompt tokens, completion tokens and how many of the prompt tokens were cache hits, which is the figure that actually decides your bill on this provider.
DeepSeek models and prices
Two models, and the relationship between them is exact: DeepSeek V4 Pro costs precisely three times DeepSeek V4 Flash on input, output and cached reads alike ($1.32 against $0.440 in, $3.96 against $1.32 out). Within each model, output costs exactly three times input. Both hold 1,000,000 tokens, so the choice is purely capability against cost.
| Model | Inputper 1M tokens | Cachedper 1M tokens | Outputper 1M tokens | Context | Counting | Your cost |
|---|---|---|---|---|---|---|
| DeepSeek V4 Pro | $1.32 | $0.044 | $3.96 | 1M | est. | — |
| DeepSeek V4 Flash | $0.440 | $0.014 | $1.32 | 1M | est. | — |
Rates in USD per 1M tokens, read from DeepSeek’s own pricing page on August 22, 2026. Providers change prices without notice.
DeepSeek’s context window
Both DeepSeek V4 Pro and DeepSeek V4 Flash accept 1,000,000 tokens — about 1,493 pages of ordinary prose at roughly 670 tokens a page. There is no smaller-window budget tier to fall back to and no larger one to upgrade into: the window is the same whichever model you pick.
Filling it is cheap by the standards of this market. 1,000,000 input tokens cost $1.32 on DeepSeek V4 Pro and $0.440 on DeepSeek V4 Flash — and if those tokens are a cache hit, $0.044. Off-peak, halve it again. Long-context work that would be a budget conversation on other providers is close to a rounding error here.
What DeepSeek tokens cost
Three request shapes on DeepSeek V4 Pro and DeepSeek V4 Flash, at standard peak rates. Every figure below halves during DeepSeek’s off-peak windows.
| Request | DeepSeek V4 Pro | DeepSeek V4 Flash |
|---|---|---|
| One chat turn1,000 in / 500 out | $0.0033 | $0.0011 |
| A long document100,000 in / 2,000 out | $0.140 | $0.047 |
| An overnight batch1,000,000 in / 50,000 out | $1.52 | $0.506 |
Two discounts stack here, and both are unusually large. Cache hits read at $0.044 per 1M against $1.32 standard — about 3.3% of the input rate, roughly a thirtieth, and the steepest cache discount among all 56 models on this site, where every other provider charges at least 10%. On top of that, DeepSeek bills 01:00–04:00 and 06:00–10:00 UTC at half price — seven hours of every twenty-four. A cached prompt sent off-peak costs a fraction of the headline number, which is why batch work is worth scheduling rather than just submitting.
DeepSeek token counter FAQ
How many tokens is 1,000 words in DeepSeek?
Roughly 1,330 tokens for English prose, and 1,000 tokens is about 752 words. DeepSeek publishes no tokenizer, so this is an approximation. Chinese text behaves differently from English — often close to one token per character — so if that is your workload, treat the estimate as a lower bound and check the usage figures the API returns.
What are DeepSeek’s off-peak hours?
01:00–04:00 and 06:00–10:00 UTC, during which DeepSeek bills at half the standard rate. That is seven hours out of every twenty-four, and DeepSeek is the only provider in this catalogue that prices by the clock at all. For anything that does not need to run immediately — evaluations, bulk summarisation, index building — scheduling into those windows halves the bill for no change in code. A 1,000,000-token prompt on DeepSeek V4 Pro costs $1.32 at peak and $0.660 off-peak.
Is this DeepSeek token counter exact?
No. DeepSeek neither publishes its tokenizer nor offers a counting endpoint, so no tool can give you an exact figure before the request runs. This counter estimates from character patterns — reliable to roughly 10–15% on English prose, less so on code and Chinese. The exact count arrives in the usage block of the API response, which also tells you how many prompt tokens were served from cache.
What is DeepSeek’s context window?
1,000,000 tokens on both DeepSeek V4 Pro and DeepSeek V4 Flash — about 1,493 pages of prose. Neither model is limited relative to the other, so you never trade window size for price within this family, which is unusual: most providers reserve their largest windows for their most expensive tier.
How much does DeepSeek V4 cost per 1M tokens?
DeepSeek V4 Pro is $1.32 in and $3.96 out; DeepSeek V4 Flash is $0.440 and $1.32. Two exact ratios make the table easy to reason about: output is three times input on both models, and DeepSeek V4 Pro is three times DeepSeek V4 Flash on every rate. Halve any of it for off-peak.
How much does DeepSeek’s cache save?
More than any other provider here. A cache hit reads at $0.044 per 1M tokens against $1.32 standard — about 3.3% of the input rate, roughly one thirtieth. Every other provider in this catalogue charges at least 10% of input for a cached read, so a stable prefix is worth far more here than elsewhere. Caching is automatic on DeepSeek rather than something you declare, so ordering your prompt with the fixed parts first is all it takes.
DeepSeek against the field
Off-peak halving and a thirtieth-price cache make DeepSeek hard to compare on headline rates alone — the effective price depends on when you run and how much you reuse. Compare 56 models from 11 providers with those caveats attached, or open another brand’s counter below.
New to tokens? Read how tokenization works