Direct answer
DeepSeek pricing has two parts. The chat app on the web, iPhone and Android is free, with no paid plan. Its API is pay as you go: deepseek-flash costs $0.15 to $0.30 per million input tokens and $0.60 to $1.20 per million output tokens, with off-peak hours at half price.
In short
- Chat app: free on chat.deepseek.com, iOS and Android. There is no Plus or Pro plan.
- API: no subscription and no tiers. You top up a balance and pay per token.
- deepseek-flash (DeepSeek-V4.1-Flash): $0.30 input and $1.20 output per million tokens at peak, half that off-peak.
- deepseek-v4-pro: $1.32 input and $3.96 output per million tokens at peak, half that off-peak.
- Cached input costs 1/50 of fresh input: $0.006 per million on Flash at peak.
- Peak hours: 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday. Weekends are off-peak all day.
- DeepSeek says unused top-up balance does not expire, and it can be refunded.
Is DeepSeek free?
Yes, for chat. The mobile app is listed as free on the App Store, and DeepSeek describes it as its official AI assistant you can use for free. The web version at chat.deepseek.com is free too. Both run the same DeepSeek-V4.1-Flash model the API sells, updated to version 1017 on 17 October 2026.
The API is not free, and the pricing docs list no free tier. Each call is deducted from a balance you top up. If DeepSeek has granted you a balance, that is spent first.
DeepSeek pricing for the API: what does each model cost?
Each model has a set price per million tokens, the same for every account. DeepSeek's help center says there is a unified pricing standard and no tiered plans. These are the list prices per million tokens, read on 21 November 2026:
DeepSeek Flash pricing (deepseek-flash)
- Input, cache hit: $0.006 at peak, $0.003 off-peak.
- Input, cache miss: $0.30 at peak, $0.15 off-peak.
- Output: $1.20 at peak, $0.60 off-peak.
- Runs DeepSeek-V4.1-Flash. A context of 1M tokens, up to 384K output tokens, vision input, tool calls and JSON output.
DeepSeek V4 Pro pricing (deepseek-v4-pro)
- Input, cache hit: $0.044 at peak, $0.022 off-peak.
- Input, cache miss: $1.32 at peak, $0.66 off-peak.
- Output: $3.96 at peak, $1.98 off-peak.
- A context of 1M tokens and up to 384K output tokens. No vision input.
Your bill is tokens times price, per model and per time window. Thinking mode is on by default, and its reasoning tokens are part of the output you pay for. As a rough guide, one English character is about 0.3 tokens.
Check the current rates on DeepSeek's Models and Pricing page.
When are DeepSeek's peak and off-peak hours?
Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday. Every other hour is off-peak, including weekends and Chinese public holidays. DeepSeek off-peak pricing is half the peak rate for every token type. The split started on 16 August 2026, with the V4 model family.
- New York (EST, winter): peak runs 8 to 11 p.m. the evening before, then 1 to 5 a.m.
- London (GMT, winter): 1 to 4 a.m. and 6 to 10 a.m.
- High-volume batch jobs, evals and backfills cost half when you schedule them outside those hours.
How does the DeepSeek cache hit price work?
A cache hit is input DeepSeek has already seen. Context caching is on for every account by default, with no code change. Say a new request starts with the same system prompt and chat history as an earlier one. That shared prefix is read from disk cache and billed at the cache hit price.
On deepseek-flash, a cache hit costs $0.006 per million tokens at peak, against $0.30 for a miss. Agent tasks and chat apps resend the same history every turn. Most of their input can then bill at the lower rate. Keep the system prompt and tool list fixed at the start of each request to get more hits.
Read how cache hits are matched in DeepSeek's context caching docs.
What will the DeepSeek API cost per month?
For most small projects, the DeepSeek API cost is a few dollars a month. The examples below use deepseek-flash at peak prices. They assume no cache hits, so they are the high end.
- Light
- 1 million input and 200,000 output tokens. $0.30 + $0.24 = $0.54.
- Medium
- 10 million input and 2 million output tokens. $3.00 + $2.40 = $5.40.
- Heavy
- 100 million input and 20 million output tokens. $30 + $24 = $54.
- The same heavy month on deepseek-v4-pro at peak
- $132 + $79.20 = $211.20.
Run the same work off-peak and each figure halves. If 80% of the input is a cache hit, the heavy Flash month drops from $54 to about $30.50 at peak.
deepseek-flash or deepseek-v4-pro: which should you pay for?
Start with deepseek-flash. It costs less than a third of Pro per token, reads images, and DeepSeek says tests by several parties put V4.1-Flash ahead of V4-Pro on performance, cost and speed. Pro is still on sale because users asked to keep it, with its billing unchanged.
- deepseek-flash
- Most chat, coding and agent work, and anything with images. Concurrency limit 2,500 requests.
- deepseek-v4-pro
- Only if your own tests show it does better on your task. Concurrency limit 500 requests.
What do DeepSeek's other APIs cost?
DeepSeek also sells search, file and voice APIs, released as a set on 13 November 2026. They bill per use, apart from tokens:
- Search API (text)
- $0.75 per 1,000 successful requests.
- Fetch API
- $0.75 per 1,000 successful requests.
- File parsing and OCR
- $0.225 per 1,000 pages.
- Text to speech
- $0.45 per 10,000 characters, $0.225 at the current 50% off.
- Speech to text
- $0.116 per hour of audio, $0.058 at the current 50% off.
Since 18 November 2026, searches the model runs during a chat call are no longer billed per call. The pages they return join the context and bill as input tokens.
How do you pay for the DeepSeek API?
You pay in advance. Top up on the DeepSeek Platform with PayPal, a bank card, Alipay or WeChat Pay. Charges show on the Billing page. The pricing docs list no monthly fees, seat prices or minimum commitments.
- Top-up balance does not expire, according to DeepSeek's FAQ.
- Unused balance can be refunded from Billing, then Refunds. Corporate bank transfers need a refund ticket.
- Invoices are requested from Billing, then Invoices.
- Need more than the concurrency limit? Ask for a rate limit increase. DeepSeek says it costs nothing extra.
- Usage can be broken down by API key on the Platform.
What happened to deepseek-chat and deepseek-reasoner?
They were the old model names, and DeepSeek set them to be discontinued on 24 July 2026. Use deepseek-flash or deepseek-v4-pro instead. The names deepseek-v4-flash and deepseek-v4-flash-vision-exp still work for now: they run on V4.1-Flash and bill at the Flash price.
That is why many DeepSeek pricing pages are out of date. Rates quoted for deepseek-chat, deepseek-reasoner or V3.2 no longer match what DeepSeek bills.
Can you use DeepSeek in Claude Code or other coding agents?
Yes. The API accepts both the OpenAI and the Anthropic request formats. In a tool that speaks the Anthropic format, such as Claude Code, set the base URL to https://api.deepseek.com/anthropic. OpenAI-format tools like OpenCode use https://api.deepseek.com. You then pay DeepSeek's token prices for that tool's model calls.
Thinking effort can be set to low, high or max. High is the default. Lower effort writes fewer reasoning tokens, so the same task costs less.
Is DeepSeek worth paying for?
For chat, there is nothing to pay for. For the API, the token prices above are the whole bill: no seat fee, no minimum spend, and off-peak hours cut them in half. Check where your data goes first. DeepSeek's privacy policy says it collects, processes and stores personal data in the People's Republic of China.
If you are choosing a chat plan rather than an API, our guides compare the paid plans of the other big assistants.
Design with DeepSeek in Claude Code
If you run Claude Code on DeepSeek, it still reads MCP servers. Better Design is an MCP server that turns a prompt into on-brand pages, slides and social images. With it, the same agent can design as well as code.
Frequently asked questions
No. The chat app is free, and the API is prepaid and billed per token. DeepSeek's help center says there are no tiered plans.
Use deepseek-flash off-peak and keep request prefixes the same so input hits the cache. Cached input then costs $0.003 per million tokens and output $0.60.
No. DeepSeek says top-up balances do not expire, and unused balance can be refunded.
Yes. The English pricing page lists every price in US dollars per million tokens or per 1,000 requests.
It limits concurrent requests per account: 2,500 on deepseek-flash and 500 on deepseek-v4-pro. Going over returns HTTP 429. Higher limits are free on request.
