LLM
Preview
Open source
DeepSeek R2
Open-weight math and science reasoning
Release2026-05
Context262K
Max output—
Parameters685B
Local / VRAM640–800 GB
LicenseOpen source
Overview
DeepSeek R2 is a model in the LLM category from DeepSeek. It was released in 2026-05. The vendor is based in China. Best for: Open-weight math and science reasoning.
Strengths
- Long chain of thought
- MIT
- Strong benchmarks
Limits
- Slow
- No vision
Specifications
| Vendor | DeepSeek |
|---|---|
| Category | LLM |
| Release | 2026-05 |
| Status | Preview |
| License | Open source |
| License name | MIT |
| Commercial use | Allowed |
| Parameters | 685B |
| Active parameters | 37B |
| Architecture | MoE reasoning transformer |
| Context window | 262K |
| Input modalities | Text |
| Output modalities | Text |
| Prompt types | Text |
| Negative prompts | No |
| Max outputs | 1 |
| Local deployment | Yes |
| VRAM | 640–800 GB |
| System RAM | 768–1536 GB |
| API access | Yes |
| Pricing | Free weights / API |
| Confidence | Medium |
| Vendor country | China |
| Vendor type | Open lab |
Pricing & hardware
Free weights / API
This model can be self-hosted or run locally. Approximate GPU memory: 640–800 GB. Approximate system RAM: 768–1536 GB. API access is available.