DeepSeek-R1

DeepSeek-R1 is positioned within DeepSeek's family of reasoning-focused models, sitting alongside base chat models like V3 and later variants such as V3.1, V3.2, and V4. A third-party guide covering the DeepSeek lineup dedicates a section to explaining R1 as a distinct member of this family, reflecting DeepSeek's approach of separating general-purpose chat capabilities from models specifically optimized for deeper reasoning tasks.

DeepSeek-R1 is described as a pioneering open-source "thinking model" that mimics human-like cognitive processes by breaking problems into smaller steps and reasoning through them logically. This reasoning-first design enables the model to tackle challenges requiring deeper analysis, such as complex logical inference and multi-step problem solving, while remaining open-weight. The model's accessible, open-source nature and reasoning orientation make it well suited for developers and researchers who need transparent access to a capable analytical model for tasks like chain-of-thought problem solving and structured reasoning workflows.

Hugging Facedeepseek-ai/DeepSeek-R1deepseek-thinking

Open provider page Provider docs

Quick Info

Provider: Hugging Face
Model key: deepseek-ai/DeepSeek-R1
Release date: Jan 20, 2025
Last updated: May 29, 2025
Knowledge cutoff: 2024-07
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.70
Output token cost: $2.50

Limits

Output tokens: 32,768 tokens
Context window: 64,000 tokens

Transparent token rates

Compare DeepSeek-R1 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about DeepSeek-R1

Together AI

Coverage

DeepSeek-R1 is framed as a January 2025 open-source reasoning model that popularized visible chain-of-thought reasoning and rivaled OpenAI's o1 at substantially lower cost. The page notes R1 has been superseded by DeepSeek V4, with the hosted deepseek-reasoner alias retired on 24 July 2026. The MIT-licensed weights remain freely available on Hugging Face for download and self-hosting, supporting offline research and reproducibility. The site's documented strengths of R1 include chain-of-thought reasoning for complex problem-solving, strong code generation across languages, and state-of-the-art mathematical reasoning benchmarks at launch.

Together AI

Coverage

DeepSeek officially announced DeepSeek-R1 on January 20, 2025, as a reasoning model positioned on par with OpenAI-o1 across math, code, and reasoning tasks. The release emphasized large-scale reinforcement learning in post-training, with the technical report hosted on the DeepSeek-R1 GitHub repository describing the approach. The launch shipped under an MIT License for both code and model weights, enabling distillation and commercial use, and DeepSeek also open-sourced six distilled variants at 32B and 70B scale aimed at OpenAI-o1-mini-class performance. API access launched via the deepseek-reasoner alias at $0.14 per million input tokens (cache hit), $0.55 per million input tokens (cache miss), and $2.19 per million output tokens.