DeepSeek API pricing: both V4 models, cheapest first
DeepSeek lists 2 DeepSeek models with a price on its official API pricing page, from DeepSeek Flash (V4.1) at $0.3 input and $1.2 output per 1M tokens to DeepSeek V4 Pro at $1.32 and $3.96 per 1M. Every rate below is checked 2026-11-18, links to its model page and names the official source.
All 2 DeepSeek models, cheapest first
| Model | Input /1M | Cached /1M | Output /1M | 1,000 calls (10K in / 1K out) | Checked |
|---|---|---|---|---|---|
| DeepSeek Flash (V4.1) | $0.3 | $0.006 | $1.2 | $0.0042 | 2026-11-15 |
| DeepSeek V4 Pro | $1.32 | $0.044 | $3.96 | $0.0172 | 2026-11-15 |
Prices are the standard list rate per 1M tokens on the official DeepSeek pricing page, checked 2026-11-18. The 1,000-call column uses a 10,000-input / 1,000-output workload with no caching; price your own token shape in the DeepSeek calculator.
What the price range means
Across the 2 DeepSeek models priced here, 1M input tokens cost $0.3 to $1.32 (a 4.4x spread) and 1M output tokens cost $1.2 to $3.96. The cheapest input rate is DeepSeek Flash (V4.1) at $0.3 per 1M; the highest is DeepSeek V4 Pro at $1.32. At a 10,000-input / 1,000-output workload the cheapest model here costs $0.0042 per 1,000 calls and the most expensive costs $0.0172.
Input and output are priced separately, so the order changes with the workload: a model with a low input rate climbs a RAG-style table, while a model with a low output rate climbs a long-answer table. The table above holds one fixed shape (10,000 in / 1,000 out) so the rows are directly comparable; price your own shape in the DeepSeek API cost calculator or the homepage calculator.
2 of the 2 models list a separate cached-input rate; where a model does not, this table reuses its standard input rate and shows a dash.
How DeepSeek bills tokens
DeepSeek publishes a per-token list price for every model in the table: input, output and, for 2 of them, a cheaper cached-input rate. The rate applies per 1M tokens; the method page explains how a price is read from the official page and what happens when one cannot be verified.
None of these models lists a batch discount on the official pricing page, so asynchronous batch work is not cheaper here.
Context windows run from 1000K tokens on DeepSeek Flash (V4.1) to 1000K on DeepSeek Flash (V4.1). A request larger than the window is rejected, so a long-document workload has to be split or moved to a larger-context model; that changes the number of calls, not just the rate.
What DeepSeek charges outside per-token pricing
DeepSeek also prices 5 services without counting tokens: search API (text), called directly, fetch API, called directly, file parsing and OCR, text to speech and speech to text. These rates are not part of the per-token tables above and do not convert into a token price, so they are listed separately, copied from the DeepSeek pricing page checked 2026-11-18.
| Service | Unit | List price | Limited-time price |
|---|---|---|---|
| Search API (text), called directly | per 1,000 successful requests | $0.75 | – |
| Fetch API, called directly | per 1,000 successful requests | $0.75 | – |
| File parsing and OCR | per 1,000 pages | $0.225 | – |
| Text to speech | per 10,000 characters | $0.45 | $0.225 |
| Speech to text | per hour of audio | $0.116 | $0.058 |
The limited-time rate is half the list price for 2 speech services (Text to speech $0.225 per 10,000 characters · Speech to text $0.058 per hour of audio). The page gives no end date for the discount, so budget at the list price and treat the difference as margin that may disappear.
Server-side search and web fetch that a model call triggers are no longer billed from 12:00 November 18, 2026 (Beijing time, UTC+8); only token usage is charged for them. Direct Search and Fetch API calls are billed at the per-request rates listed in the table.
For scale: 1,000 parsed pages cost $0.225, and one hour of audio costs $0.116 to transcribe. A bill that hits the per-page or per-request rates above carries both parts, the token rows above and the per-call charges here.
DeepSeek pricing page, checked 2026-11-18.
Which DeepSeek model to price first
Start with the cheapest model that passes your own test, not with the most expensive one. DeepSeek Flash (V4.1) has the lowest input rate in this lineup at $0.3 per 1M tokens; if its answers hold up on your prompts, moving a workload down to it is usually worth more than any caching or batch tweak. If it fails, step up one row at a time in the table above and re-run the same prompts, so the quality difference you pay for is one you can see.
Cost per call is not the same as cost per finished task: a cheaper model that needs two attempts, a longer prompt or manual fixes costs more than its rate suggests. The benchmark comparison puts measured scores next to these prices, and the coding-cost ranking prices completed coding tasks rather than single calls.
Related
- DeepSeek API cost calculator
- Every model price per 1M tokens
- Cheapest LLM APIs, ranked by cost per call
- Models grouped by price tier
- Blended cost by input:output mix
- DeepSeek V4 Pro vs GPT-6.1 Sol
- OpenAI models and prices
- Anthropic models and prices
- Google models and prices
- xAI models and prices
- Mistral models and prices
Questions
How many DeepSeek models does this page price?
This page prices 2 DeepSeek models, every one copied from the official DeepSeek pricing page and checked 2026-11-18. Each row links to a per-model page with the full rate tiers, worked examples and the source URL, so a number here can always be traced back.
Does DeepSeek charge less for cached input tokens?
Yes. 2 of the 2 models list a cached-input rate below the standard input rate; the cheapest cached read here is $0.006 per 1M tokens. Cached pricing only applies to a repeated prefix the provider has stored, so the saving depends on how much of your prompt is identical between calls.
Which DeepSeek model is cheapest per 1M tokens?
DeepSeek Flash (V4.1) has the lowest input rate at $0.3 per 1M tokens and $1.2 per 1M output tokens. At a 10,000-input / 1,000-output workload the cheapest model is DeepSeek Flash (V4.1) at $0.0042 per 1,000 calls, while DeepSeek V4 Pro at $1.32/$3.96 carries the highest input rate in this lineup.
Which DeepSeek model has the largest context window?
The largest context window here is 1000K tokens on DeepSeek Flash (V4.1). A larger window does not make a model cheaper: cost the input side of the bill first and check the window only for the documents you need to send in one request.
Next step → Price a DeepSeek workload per user
Price data last checked 2026-11-18 against official sources · how we verify
Price-change alerts
One short email when a provider changes a price or a promo ends. No spam, unsubscribe any time.