Skip to main content
Rase AI Intelligence Newsletter

Models & assistants Model or AI service

DeepSeek released R1 reasoning weights for local use

DeepSeek released its R1 reasoning model in January 2025 with downloadable weights. It also published six smaller models trained on examples generated by R1. [1][2][4]

Model access
Original R1 weights remain downloadable. DeepSeek’s hosted launch used an identifier later assigned to newer models. [1][2][4]
License
MIT for original R1 weights. Consult the model card for upstream terms of distilled variants. [4]

What the model does

The January 20 release included original R1 weights and hosted access at launch. DeepSeek later assigned the same hosted model name to newer models. [1][2]

Why it matters

Researchers and developers can keep a fixed downloadable checkpoint for repeated tests. That helps when a hosted endpoint changes which model it runs. [1][4]

Who can use it

Teams evaluating models for mathematics, coding and reasoning can compare the original checkpoint with smaller variants. They have different sizes and base models. [4]

Key dates

  1. Original DeepSeek-R1 release [1][2]

Ways to use it

A developer testing code explanations can record the exact checkpoint in an evaluation so that a different endpoint or distilled variant does not silently change what is being compared. [1][4]

Model limitations

The original R1 checkpoint is much larger than the distilled alternatives. DeepSeek’s model card identifies the base-model licenses relevant to the smaller variants. [4]

Sources and verification

Catalog evidence checked How we verify records

  1. Primary source DeepSeek’s official API release history
  2. Primary source DeepSeek-R1 Release | DeepSeek API Docs
  3. Independent source [2501.12948] DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
  4. Independent source deepseek-ai/DeepSeek-R1

Related profiles

Models & assistants

GPT-6.2 Astra and Sol

OpenAI

Astra targets demanding projects. Sol targets everyday tasks. Both support text and image input in the API.

High confidenceEvidence checked

Models & assistants

MAI Code 2 and Code 2 Flash

Microsoft AI

Code 2 targets complex software work. Flash targets frequent edits, explanations and routine fixes.

High confidenceEvidence checked

Models & assistants

MAI Realtime 1

Microsoft AI

Users can try a MAI Labs demo. Developers can register for a limited Foundry preview.

High confidenceEvidence checked