DeepSeek Pricing in 2026: The App Is Free, the API Is Not
Every "DeepSeek pricing" article I could find starts by comparing subscription tiers. There are none. DeepSeek's homepage has no upgrade button and no plan table — the only price link on it goes to API pricing. So the real question is not "which plan", it is "where does free stop". Here are the exact numbers.
Short answer: there is no DeepSeek subscription. The website and app are free, with no paid tier to upgrade to. The only thing DeepSeek sells is API access, billed per million tokens: as of 2 November 2026 that is $0.003–$0.30 per million input tokens and $0.60–$1.20 per million output tokens on the cheap deepseek-flash model, depending on whether you hit the cache and whether you are inside the peak window. A normal question-and-answer exchange costs well under one cent.
1. What is actually free
This is the part that gets lost. We opened deepseek.com on 2 November 2026 and read the whole page. The product links are: the app download, the web version, DeepSeek Harness (a desktop client), the API platform, the API docs, API pricing, and service status. There is no "Pricing" link for the consumer product, no "Upgrade", no plan comparison, no credit card prompt.
That absence is the answer. Compare it with ChatGPT, Gemini or Claude, where the pricing page is one of the most-linked pages on the domain. If DeepSeek had a paid consumer tier, it would be on the homepage.
| What you use | What it costs | How we know |
|---|---|---|
| Web chat (chat.deepseek.com) | Free, no plan to upgrade to | Our own free account, no payment details entered |
| Mobile app | Free | Same account, sign-in carries over |
| API access | Paid, per million tokens | Rates quoted below, from the official docs page |
| Search / Fetch API | Paid since 1 Nov 2026, $0.75 per 1,000 successful requests | DeepSeek pricing page, billing note |
We have run a free DeepSeek account since October 2026 and have never entered payment details. In our comparison test we sent 10 rapid messages and 5 image uploads in a row and hit no limit popup, no cooldown and no upsell — the behaviour is documented in our DeepSeek vs ChatGPT test. Third-party sites describe the web version as "completely free and uncapped"; we could not find evidence for the "uncapped" half beyond our own run, and 15 requests is not enough to prove a cap does not exist at some higher volume.
2. The API rates, exactly as published
DeepSeek publishes two models, and every price is per one million tokens. A token is the smallest unit the model reads — roughly three quarters of an English word, so 1,000 words is about 1,300 tokens.
| Per 1M tokens | deepseek-flash off-peak | deepseek-flash peak | deepseek-v4-pro off-peak | deepseek-v4-pro peak |
|---|---|---|---|---|
| Input, cache hit | $0.003 | $0.006 | $0.022 | $0.044 |
| Input, cache miss | $0.15 | $0.30 | $0.66 | $1.32 |
| Output | $0.60 | $1.20 | $1.98 | $3.96 |
Three things about this table are easy to miss:
- Output costs four times more than input on a cache miss. $0.60 against $0.15 for flash. If you are optimising a bill, shortening the model's answers saves far more than shortening your prompts.
- A cache hit is 50 times cheaper than a cache miss. $0.003 against $0.15. That is why anything that reuses a long system prompt benefits enormously from prompt caching — the repeated opening of every request lands in the cheap column.
- Peak doubles everything. Not a small premium; a clean 2x on every line of the table.
Both models carry a 1M token context window and up to 384K tokens of output. The concurrency limit is 2,500 for flash against 500 for v4-pro. Model names matter: deepseek-flash maps to DeepSeek-V4.1-Flash-1017, and the older names deepseek-v4-flash and deepseek-v4-flash-vision-exp still resolve but are now served by the V4.1 model at the Flash price.
One capability difference is worth knowing before you pick: vision is supported on flash and not on v4-pro. If your input includes images, that decision is made for you.
3. The half-price window, in your timezone
DeepSeek's peak hours are defined in UTC, which is unhelpful if you are not. Converting:
| Zone | Peak (2x price) | Off-peak (half price) |
|---|---|---|
| UTC | 01:00–04:00 and 06:00–10:00 | all other hours |
| Beijing (UTC+8) | 09:00–12:00 and 14:00–18:00 | 18:00–09:00 |
| New York (UTC−5, winter) | 20:00–23:00 and 01:00–05:00 | 05:00–20:00 |
| London (UTC+0, winter) | 01:00–04:00 and 06:00–10:00 | all other hours |
| Sydney (UTC+11) | 12:00–15:00 and 17:00–21:00 | 21:00–12:00 |
Peak applies Monday to Friday only, and excludes Chinese public holidays — and on those holidays the whole day bills at the off-peak rate, not just the peak hours. Weekends are entirely off-peak.
There is an irony in the Beijing row: the two peak blocks are exactly Chinese office hours. The cheapest time for a team in China to run a batch job is late evening or before 9am, which is the opposite of what most people assume about a Chinese provider.
4. What one conversation costs
Per-million-token pricing is hard to feel. Here is the arithmetic, worked through at peak rates on the cheaper model, which is the worst case for a normal exchange.
Say you type a 200-word question and get a 500-word answer. At roughly 1.33 tokens per word that is about 270 input tokens and 670 output tokens — and because your question has not been seen before, it is a cache miss.
| Line | Tokens | Rate (flash, peak) | Cost |
|---|---|---|---|
| Your question, cache miss | 270 | $0.30 / 1M | $0.000081 |
| The answer | 670 | $1.20 / 1M | $0.000804 |
| One exchange | 940 | — | ≈ $0.0009 |
| 1,000 exchanges | 940,000 | — | ≈ $0.90 |
| Same, off-peak | 940,000 | half | ≈ $0.45 |
Under a tenth of a cent per exchange. A thousand of them — which is more conversation than most people have with any chatbot in a year — lands under a dollar.
If you want this done against your own traffic rather than one example, the DeepSeek API cost calculator takes your request volume, token counts, cache hit rate and peak share and returns a monthly figure, plus a worked comparison of which cost lever is worth pulling first.
Long documents change the shape of this. If you paste a 50-page PDF of about 25,000 words, that is roughly 33,000 input tokens, and it goes in on every turn of the conversation. On flash at peak that is $0.01 per turn for the document alone. Ten follow-up questions against the same PDF costs about ten cents. On v4-pro at peak the same thing is $0.044 per turn — which is exactly the gap that prompt caching exists to close.
5. DeepSeek vs ChatGPT Plus, Gemini and Claude
The comparison only works if you compare the same thing. These are two different products wearing the same word, "price":
| DeepSeek | ChatGPT | Gemini | |
|---|---|---|---|
| Consumer chat price | $0 — no paid tier exists | $20/mo (Plus) | $0, with a $4.99/mo limit-doubling add-on |
| What the paid tier adds | nothing to buy | higher capacity, extra features | double the daily and weekly allowance |
| API billing | per token, from $0.003/1M | separate API account | separate API account |
| Free tier limits we hit | none in 10 messages + 5 uploads | popups reported by users; uploads have separate unpublished limits | visible daily and weekly meters |
The ChatGPT and Gemini figures are from our own test day in October 2026: ChatGPT Plus at $20/month, shown with a first-month promo at $10 and fine print stating the price continues at $20 from 25 November 2026; Gemini's usage page showed a daily bar, a weekly bar and a $4.99/month offer to double both. Full detail in the three-way test.
The honest summary: for a person who just wants to ask questions, DeepSeek is the only one of the three where the price question never comes up. Not because it is cheaper — because there is nothing to buy. Whether that is worth more than ChatGPT's other features is a different article; on price alone, it is not close.
6. Search and Fetch started charging on 1 November
This one is new enough that most write-ups predate it. DeepSeek's Search and Fetch APIs are billed per successful request at $0.75 per 1,000 — about 0.075 cents each. Failed requests are not charged.
The billing note adds a detail worth knowing if you use a coding agent: server-side tool calls made through DeepSeek Harness, Claude Code or Codex used to hit these APIs for free, and per-request billing started at 00:00 Beijing time on 1 November 2026.
So an agent that searches the web twenty times while working on your task now adds about 1.5 cents to the run. Trivial per run, less trivial across thousands.
7. When paying makes sense — and when it does not
Because there is no subscription, the decision is not "free vs paid". It is "do I need an API key at all".
| Your situation | Verdict |
|---|---|
| You type questions into a chat box | Pay nothing. There is no tier above you and no feature you are locked out of by not paying. |
| You want DeepSeek inside your own script, app or automation | Pay per token. At these rates the bill is unlikely to be the thing that stops you — a $5 top-up covers thousands of exchanges. |
| You are processing long documents repeatedly | Turn on prompt caching and schedule for off-peak. Those two moves cut a bill by far more than switching models would. |
| You mostly generate output rather than read it | Watch output length, not input. Output is 4x the input rate on a cache miss; a verbose answer is where the money goes. |
| You need image input on the pro model | You cannot. Vision is flash-only. Not a price issue, but it decides the model for you. |
And the thing nobody selling you a plan will say: if the free web app already does what you need, an API key adds nothing but a bill. The API is for programmatic access. If you are not writing code, you are not the customer for it.
Frequently asked questions
- Is DeepSeek free?
- The chat app and website are free, and there is no paid subscription to upgrade to — DeepSeek's homepage lists no plan at all. What costs money is the API, billed per million tokens.
- How much does the DeepSeek API cost?
- As of 2 November 2026,
deepseek-flashis $0.003 per million input tokens on a cache hit, $0.15 on a cache miss, and $0.60 per million output tokens off-peak — double at peak.deepseek-v4-prois $0.022, $0.66 and $1.98, also doubling at peak. - When are DeepSeek's peak hours?
- 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays. In Beijing time that is 09:00–12:00 and 14:00–18:00. Weekends and holidays are off-peak all day.
- What does one conversation cost on the API?
- About $0.0009 at peak on flash for a 200-word question and a 500-word answer — under a tenth of a cent. A thousand such exchanges is roughly $0.90, or $0.45 off-peak.
- Does DeepSeek give free API credits?
- No free allowance is advertised on the pricing page as of 2 November 2026. Usage is deducted from a topped-up balance or a granted balance, with the granted balance used first when both exist. We have not opened a paid API account, so top-up minimums are untested by us.
- Are the Search and Fetch APIs free?
- No longer. They are $0.75 per 1,000 successful requests, and per-request billing for server-side tool calls through Harness, Claude Code and Codex began at 00:00 Beijing time on 1 November 2026.
- Is DeepSeek cheaper than ChatGPT Plus?
- They are not the same product. ChatGPT Plus is a $20/month chat subscription; DeepSeek's chat app is free with no subscription at all. A heavy individual chat user pays $0 on DeepSeek. A developer pays per token on DeepSeek and needs a separate API plan from OpenAI.
Sources and what we could not test
| Claim | Where it came from | Date checked |
|---|---|---|
| All per-million-token rates for deepseek-flash and deepseek-v4-pro, peak and off-peak | DeepSeek API docs — Models & Pricing, pricing table | 2 Nov 2026 |
| Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Mon–Fri, excluding Chinese public holidays; off-peak is half price | DeepSeek API docs — Models & Pricing, footnote 2 | 2 Nov 2026 |
| 1M context, 384K max output, concurrency 2,500 / 500, vision on flash only | DeepSeek API docs — Models & Pricing, model details table | 2 Nov 2026 |
| Search and Fetch at $0.75 per 1,000 successful requests; billing for agent tool calls began 00:00 Beijing time 1 Nov 2026 | DeepSeek API docs — Search & Fetch Pricing | 2 Nov 2026 |
| No consumer pricing, plan table or upgrade prompt anywhere on the homepage | deepseek.com, full page read | 2 Nov 2026 |
| Free account: no limit prompt across 10 rapid messages and 5 image uploads | Our own free DeepSeek account, no payment details entered | Oct 2026 |
| ChatGPT Plus $20/mo with $10 first-month promo; Gemini $4.99/mo limit doubling | Our own accounts, upgrade pages as displayed | Oct 2026 |
| Token-per-word and cost-per-conversation arithmetic | Derived from the published rates above; not a received invoice | 2 Nov 2026 |
What we could not test: whether the free web app has any cap at very high volume (we ran 15 requests, not 1,500); the minimum API top-up and whether a granted balance is offered to new accounts; the actual billed amount for a real conversation, since we did not fund an API account; and how quickly prompt caching actually kicks in on a repeated long prompt. The cost figures in section 4 are arithmetic, clearly labelled as such, not a bill.