DeepSeek
DeepSeek is an AI lab that builds open-source large language models and reasoning models, released openly under permissive licenses.
Our verdict
Who it's for
Open-source large language and reasoning models, efficiently trained
Who should look elsewhere
Server capacity issues during peak times
Pros
- +Significantly lower costs compared to leading models
- +Open-source MIT-licensed models for self-hosting
- +Excellent performance on coding and math tasks
- +Chain-of-thought reasoning for complex problems
- +No vendor lock-in with self-hosting option
Cons
- −Server capacity issues during peak times
- −Data stored in China may raise privacy concerns
- −Weaker at creative writing compared to competitors
- −Occasional censorship on sensitive topics
Key features
- ✓DeepSeek-V3 and R1 large language models
- ✓Chain-of-thought visible reasoning
- ✓64K token output support
- ✓API access with pay-per-token pricing
- ✓Self-hosting with MIT license
- ✓Code generation and debugging
- ✓Mathematical problem solving
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| Free | $0 | month | Basic API access; Rate limited |
| Pay-as-you-go | $0.07 | month | $0.07/M input (cache hit); $0.56/M input (cache miss); $1.68/M output tokens |
| On-premise | $18000 | month | Full infrastructure control; No API limits; Data stays on-premise |