A free chat assistant from the Chinese lab behind the DeepSeek open-weight models.
- Starting price
- Free to use; API from $0.14 per 1M tokens
- Pricing model
- Free
- User rating
- 4.5 / 5
- Best for
- Free chat and very cheap API access
DeepSeek is a Chinese AI lab whose models undercut Western rivals on price by a wide margin. The chat app is free, and the API costs a fraction of what OpenAI or Anthropic charge.
Who DeepSeek is for
Two groups. People who want a capable assistant without a subscription, and developers who need volume at low cost.
The trade-off is data residency. Queries go to servers in China, which rules it out for a lot of business use. Read the terms before putting anything sensitive through it.
What DeepSeek costs
The chat interface is free with no paid consumer tier. There is nothing to buy and no usage plan to pick.
The API is billed per token. V4 Flash runs at $0.14 per million input tokens and $0.28 per million output. V4 Pro is $1.74 and $3.48, with promotional discounts often bringing that down sharply.
Prefix caching cuts repeat input costs by around 98%, which matters a lot if you send the same context repeatedly.
DeepSeek pricing and plans
Chat app
$0
free to use
No paid consumer tier exists.
- Web, iOS and Android
- Reasoning models included
- Web search
- No message subscription
API — V4 Flash
$0.14
per 1M input tokens
Output $0.28 per 1M. Cache hits $0.0028.
- 1M token context window
- 384K maximum output
- Automatic prefix caching
- Best price for high volume
API — V4 Pro
$1.74
per 1M input tokens
Output $3.48 per 1M. Cache hits $0.0145.
- Stronger reasoning
- 1M token context window
- Promotional discounts often apply
- Automatic prefix caching
Pricing checked against the official DeepSeek pricing page on 5 September 2026. Vendors change plans often, so confirm before you buy.
What DeepSeek actually does
- Free chat on web, iOS and Android
- Reasoning models at no cost to consumers
- 1M token context window on current models
- Automatic prefix caching, around 98% off repeat input
- Open weights for several models
- OpenAI-compatible API format
- Web search inside the chat app
Pros and cons
What works well
- Chat app is completely free with no message subscription
- API pricing undercuts every major Western lab
- 1M token context window on current models
- Prefix caching cuts repeat costs by around 98%
- Open weights available for self-hosting
What to watch out for
- Data is processed on servers in China
- No permanent API free tier for production
- Very small third-party review sample
- No paid consumer tier if you want guaranteed capacity
DeepSeek FAQs
Is DeepSeek free?
The chat app is free on web and mobile, with no paid consumer tier. The API is billed per token.
How much does the DeepSeek API cost?
V4 Flash is $0.14 per million input tokens and $0.28 per million output. V4 Pro is $1.74 and $3.48. Cached input is around 98% cheaper.
Is there an API free tier?
New accounts get promotional credits, but there is no permanent free tier for production workloads.
Is DeepSeek safe to use for work?
Queries are processed on servers in China. That is a data residency problem for many organisations, so check your own policy before using it with company information.
How does it compare with ChatGPT on price?
It is dramatically cheaper. The consumer app costs nothing, and API rates are a small fraction of the major Western labs for comparable context windows.
Can I run DeepSeek models myself?
Yes. Several DeepSeek models have open weights and can be run on your own hardware or through third-party hosts.
What is prefix caching?
The API automatically reuses identical leading context between calls and charges roughly 98% less for it. No code changes are needed.
Where does the 4.5 rating come from?
It is the aggregate score from 20 reviewers on G2, checked 12 August 2026. The sample is small, so treat it as a weak signal.
Weighing up other options? Read our roundup of AI tool alternatives.
Ready to try DeepSeek?
Go to DeepSeek