DeepSeek LLMs
Chinese lab known for high-performance open-weight models at low prices.
DeepSeek-V3 and the R1 reasoning model upended the price-performance frontier in early 2026, offering near-GPT-4 capability at a fraction of the per-token cost. Aggressive prompt-caching discounts make it particularly attractive for high-volume coding workloads.
Founded: 2023 · HQ: Hangzhou, China · Docs: platform.deepseek.com ↗
All DeepSeek models
| Model | Family | Context | Input $/M | Output $/M | Released | Status |
|---|---|---|---|---|---|---|
| DeepSeek-R1 | DeepSeek-R1 | 128K | $0.55 | $2.19 | 2025-01-20 | active |
| DeepSeek-V3 | DeepSeek-V3 | 128K | $0.27 | $1.10 | 2024-12-26 | active |
Comparisons with other providers
The most-searched comparisons involving DeepSeek models:
- DeepSeek-R1 vs GPT-5 Nano
- DeepSeek-R1 vs GPT-4o Mini
- DeepSeek-V3 vs GPT-5 Nano
- DeepSeek-V3 vs GPT-4o Mini
Working with the DeepSeek API
Documentation lives at https://platform.deepseek.com/docs. Before paying for any call, count input tokens with the appropriate counter:
- OpenAI Token Counter — for GPT-* and o-series models
- Claude Token Counter — for Anthropic Claude models
- Gemini Token Counter — for Google Gemini models