DeepSeek is a Chinese research company whose assistant became widely known for delivering strong reasoning and coding performance at a fraction of the usual API cost, with model weights published under permissive terms so organisations can run them on their own hardware. The consumer chat application is free and includes a visible reasoning mode and web search. For developers, the API is priced well below Western equivalents and is broadly compatible with existing client libraries, which makes switching cheap to test. It suits cost-sensitive deployments, teams that need self-hosted models for data residency reasons, and developers experimenting with reasoning models. The limitations are governance rather than capability: data submitted to the hosted service is processed in China and subject to local law, the models decline or deflect on topics sensitive to that jurisdiction, and some organisations and governments restrict the app on managed devices, so check policy before adopting it at work.
How to use DeepSeek
1The chat application at deepseek.com works with an email or phone sign-up and is free to use, including the reasoning mode that shows its intermediate thinking before answering.
2For development, create an API key in the platform console and top up a balance, since billing is prepaid rather than invoiced; the client interface follows the widely used chat completion format, so most existing code works by changing the base URL and key.
3Choose between the general and reasoning models explicitly, as the reasoning model costs more and is slower but is the reason to use the service.
4For self-hosting, download the published weights and serve them with a standard inference runtime, which requires substantial GPU memory for the larger variants.
5The considerations that matter most are not technical: confirm your organisation permits the hosted service before sending anything confidential, and expect the models to avoid politically sensitive subjects.
6For production use, set a spending cap on the account, because prepaid balances can drain faster than expected under load.
Best for
cost-sensitive API use
self-hosted deployment
reasoning tasks
coding assistance
experimentation
Pros and cons
What it does well
Very low API pricing compared with competitors.
Model weights published for self-hosting.
Visible reasoning mode in the chat application.
Compatible with common client libraries.
Where it falls short
Hosted data is processed under Chinese jurisdiction.
Models avoid politically sensitive topics.
Restricted on managed devices in some organisations.
DeepSeek pricing
Free tierPrices as listed by the maker; check the site for the latest.
Free chat application
usage-based API pricing that is substantially cheaper than most competitors
open weightsreasoning modellow cost apiself hostingcoding