DeepSeek
Chinese AI lab delivering frontier-level reasoning at open-source pricing.
About
About DeepSeek
DeepSeek, backed by the Chinese quantitative hedge fund High-Flyer, shocked the AI industry in early 2025 by releasing DeepSeek-R1 — a reasoning model that matches o1 performance while being available as open weights. DeepSeek's models use Multi-head Latent Attention (MLA) and mixture-of-experts architecture for exceptional efficiency. API access is priced 10–30× cheaper than comparable Western frontier models.
Key highlights
DeepSeek-R1 matches OpenAI o1 on AIME 2024 and MATH-500 — released as open weights
API pricing 10–30× cheaper than comparable Western frontier models
MIT-licenced open weights for V3 and R1 — full commercial use permitted
Multi-head Latent Attention (MLA) architecture for exceptional inference efficiency
Caused significant market impact in 2025, demonstrating compute efficiency
Models
3 models tracked
Context windows, pricing, capabilities, and API identifiers — everything you need to choose the right model for your workload.
DeepSeek-V3
2024
DeepSeek's flagship general-purpose model. A 685B MoE architecture with 37B active parameters. Available via API at a fraction of GPT-4 pricing or as open weights for self-hosting.
- Extreme cost efficiency
- Strong coding quality
- Open weights (MIT)
- 685B MoE architecture
deepseek-chatDeepSeek-R1
2025
An open-weight reasoning model that matches OpenAI o1 performance on math and coding benchmarks. Uses reinforcement learning for chain-of-thought — released as MIT-licenced open weights.
- Matches o1 on AIME 2024
- Extended chain-of-thought
- MIT-licenced open weights
- RL-trained reasoning
deepseek-reasonerDeepSeek-V2.5
2024
An updated version of DeepSeek V2 combining chat and coding. Offers 128K context and is widely supported across open-source inference frameworks.
- 128K context
- Web + code capabilities
- Open weights
- Wide inference support
deepseek-v2.5Use Cases
What DeepSeek is best for
Each provider has specialisms shaped by their architecture decisions, training data, and API design. Here is where DeepSeek performs best.
Get started
Start building with DeepSeek
Visit the official DeepSeek platform to access API keys, documentation, and pricing details.