LLMOpen weights
DeepSeek-R1
DeepSeek
Updated May 21, 2026 · Benchmarks via Artificial Analysis, specs & pricing via OpenRouter · Methodology · Report an error
Specifications
- Type
- LLM
- Access
- Open weights
- Released
- January 20, 2025
- License
- MIT
- Parameters
- 671B total / 37B active (MoE)
- Context window
- 128,000 tokens
- Output speed
- 189 tok/s
- Latency (TTFT)
- 0.07s
- Input
- Text
- Output
- Text
- API pricing
- $0.55 in · $2.19 out / 1M tokens
Capabilities
Function callingStructured output
Benchmarks
Our take
The first open-weights model to rival OpenAI’s o1 on reasoning — released under a permissive MIT license. Trained largely with reinforcement learning, it stunned the field on capability and cost and triggered a market-wide rethink of closed reasoning models.