# DeepSeek-R1

> Open-source reasoning model rivaling proprietary models in math, code, and logic

DeepSeek-R1 is an open-source reasoning model developed using reinforcement learning to generate explicit chain-of-thought outputs. Built on a mixture-of-experts architecture with 671B total parameters (37B active), it performs on par with proprietary reasoning models like OpenAI o1. It is accessible for free via DeepSeek's web chat and app, available via API, and released under an MIT license alongside smaller distilled variants.

Website: https://huggingface.co/deepseek-ai/DeepSeek-R1

## What it is for

- Solve multi-step mathematical problems and formal proofs
- Debug and generate complex software code across various programming languages
- Perform deep logical reasoning and step-by-step analytical reasoning tasks
- Distill reasoning datasets to train smaller, lightweight open-source models

## Features

- Reinforcement learning-driven chain-of-thought reasoning with internal reflection tokens
- 671B total parameter mixture-of-experts (MoE) architecture with 37B active parameters
- 128k context length with output support up to 32k tokens
- Open MIT license permitting commercial use, fine-tuning, and data distillation
- Distilled smaller models available in 1.5B, 7B, 8B, 14B, 32B, and 70B sizes
- OpenAI-compatible API support via DeepSeek Platform

## Pricing

Open source. Free to use via web chat and mobile app; pay-as-you-go API; model weights are free under MIT license.

## Details

- Type: AI model
- Category: [AI](https://xiry.app/topics/ai.md)
- Open source: yes
- Platforms: Web, iOS, Android, API, Self-hosted
- For: Developers, AI researchers, and students needing deep reasoning capabilities for math, code, and logic.
- Alternative to: OpenAI o1, OpenAI o1-mini, Claude 3.5 Sonnet
- Tags: deepseek, reasoning, llm, open source, mixture of experts, coding, mathematics

## Why it was saved

- [@JulianGoldieSEO, 27 January 2025](https://x.com/JulianGoldieSEO/status/1883732510938640528): The tweet promises a tutorial on automating anything with it; free access via deepseek.com.

## Related

- [DeepSeek](https://xiry.app/resources/deepseek): Open-weights AI models and chat interface for reasoning, coding, and research. (AI model, Freemium)
- [DeepSeek-V3](https://xiry.app/resources/deepseek-v3): Open-weight 671B mixture-of-experts language model matching proprietary frontier systems (AI model, Open source)
- [DeepSeek Chat](https://xiry.app/resources/deepseek-chat): Conversational AI assistant with open-weights reasoning and general language models (Web app, Freemium)
- [Pinokio](https://xiry.app/resources/pinokio): Browser that installs and runs open-source AI apps with one click (Desktop app, Free)
- [Claude](https://xiry.app/resources/claude): Anthropic's conversational AI assistant for analysis, writing, and coding (Web app, Freemium)
- [Hermes 3](https://xiry.app/resources/hermes-3): Open-weights fine-tune of Llama 3.1 built for agentic function calling. (AI model, Open source)

Updated 10 October 2026. Page: https://xiry.app/resources/deepseek-r1

Make your own library: Xiry takes your saved reels, tweets, Reddit bookmarks and Watch Later videos, makes them searchable and pulls the useful stuff out, like this library. It's invite-only for now; join the waitlist at https://xiry.app/#waitlist.