AI Radar

DeepSeek-R1

Available · Language, Reasoning · Milestone

DeepSeek-R1 is a reasoning language model built on DeepSeek-V3-Base, released on 20 January 2025 in the DeepSeek web chat (DeepThink), in the API as deepseek-reasoner and as open weights under the MIT License. DeepSeek trained it in two fine-tuning and two reinforcement learning stages to address the repetition, readability and language-mixing problems of DeepSeek-R1-Zero. [1] [2] Primary source

Timeline of DeepSeek-R1 →

Claims and evidence

  • Released Primary source[1]news sidebar: DeepSeek-R1 Release 2025/01/20 [3]Date: 2025-01-20 [4]page date
  • Status AvailableWeights are still public on Hugging Face. In the DeepSeek API, deepseek-reasoner moved to DeepSeek-R1-0528 on 2025-05-28. Primary source[2] [3]Date: 2025-05-28
  • Replaced by DeepSeek-R1-0528In the DeepSeek API only: deepseek-reasoner was upgraded to DeepSeek-R1-0528. The weights stay available. Primary source[3]Date: 2025-05-28
  • Derived from (fine tune) DeepSeek-V3Trained from the base variant DeepSeek-V3-Base: fine-tuned on cold-start data, then two reinforcement learning stages and two supervised fine-tuning stages in total. Primary source[2]2. Model Summary; 3. Model Downloads [5]Introduction
  • Successor of DeepSeek-R1-Lite-Preview EditorialEditorial link along the R line: DeepSeek-R1-Lite-Preview was DeepSeek's earlier reasoning preview; the R1 announcement does not mention it. Primary source[1]
  • Change · Availability Offered in the API and as open weights, while DeepSeek-R1-Lite-Preview was available only in the DeepSeek web chat at launch.Compared with DeepSeek-R1-Lite-Preview Primary source[1] [6]
  • Change · Licensing Code and weights under the MIT License, and DeepSeek allows the weights and API outputs to be used for fine-tuning and distillation.Compared with DeepSeek-R1-Lite-Preview Primary source[1]License Update
  • Open weights YesMIT License; DeepSeek allows using the weights and API outputs for fine-tuning and distillation. Primary source[1]License Update [2]7. License
  • Context window 128K tokensModel context length per the downloads table; API limits may differ. Primary source[2]3. Model Downloads [7]3. Model Downloads
  • Parameters 671B total, 37B activatedMixture-of-experts. Primary source[2]3. Model Downloads [7]3. Model Downloads
  • Access consumer app, API, Open-weights download Primary source[1]first section; API Access & Pricing [2]5. Chat Website & API Platform

Lineage

Predecessors

Successors

No known successor.

Based on

  • DeepSeek-V3 · 26 December 2024 · derived (fine tune)

Variants and derived

Siblings

None recorded.

All ancestors

All descendants

Variants

No variants recorded in this record.

Related AI Radar coverage

AI Radar coverage starts in June 2026; no coverage linked yet.

All model releases from DeepSeek on AI Radar →