Plain-English Summary
A widely used open reasoning model designed to show strong step-by-step performance in math, coding, and logic.
DeepSeek-R1
A widely used open reasoning model designed to show strong step-by-step performance in math, coding, and logic.
DeepSeek R1
Reasoning LLM
Available / open weights
Open weights: Yes
API: Yes
Technical Notes
671B-parameter MoE model with about 37B active parameters, long context, reinforcement-learning-based reasoning, and distilled variants.
Best For
Math, coding, reasoning research, self-hosted assistants
Watch Out For
Reasoning can be verbose; large model requires significant hardware and may need safety tuning.
Model Specs
- Maker
- DeepSeek
- Country
- China
- Family
- DeepSeek R1
- Type
- Reasoning LLM
- Release date
- 2025-01-20
- Date confidence
- Exact
- Status
- Available / open weights
- Context window
- 128,000 tokens
- Max output
- 64,000
- Parameters
- 671B
- Active parameters
- 37B
- Architecture
- Mixture-of-experts transformer
- Reasoning
- Yes — reasoning model
- Tool calling
- Model / host dependent
- Structured output
- Model / host dependent
- API available
- Yes
- Open weights
- Yes
- Self-hostable
- Yes
- License
- MIT
- Inputs
- Text
- Outputs
- Text
- Approx. price
- See provider pricing
- How to access
- Provider website or API
- Model / API ID
- deepseek-reasoner / DeepSeek-R1
- Knowledge cutoff
- Not publicly disclosed
- Fine-tuning
- Fine-tuning, distillation, self-hosting
- Completeness
- 100 %
- Verified
- Jul 18, 2026
- Snapshot date
- Jul 18, 2026
- Source
- Official source