Model
DeepSeek-R1
DeepSeek-R1 is an open-weight reasoning model that DeepSeek released on 20 January 2025. Built on DeepSeek-V3-Base and trained with reinforcement learning, it is released under the MIT License, which allows commercial use and distillation.
Status DeprecatedModality TextOpen weights YesContext (K tokens) 128 Era Generative AI for everyone
- Parameters
- 671B
- License
- MIT
- Access
- API, App, Download
- Release date
- Developer
- DeepSeek
What it is
DeepSeek-R1 is an open-weight reasoning model that DeepSeek released on 20 January 2025. Its technical report followed on arXiv on 22 January 2025. [7][1]
It is built on DeepSeek-V3-Base, which has 671 billion total parameters, 37 billion activated per token and a 128K-token context. [1]
Origins
The series began with DeepSeek-R1-Lite-Preview, launched on chat.deepseek.com on 20 November 2024. That preview was described with o1-preview-level results on AIME and MATH and showed the model's real-time thought process. [6]
DeepSeek-R1-Zero was trained from the base model with reinforcement learning alone, without supervised fine-tuning as a first step. It developed sophisticated reasoning behaviours without being explicitly taught how to reason. R1 added cold-start data and a multi-stage pipeline to fix R1-Zero's poor readability and language mixing. [1]
Nature's supplementary information gives the training cost of R1 as 147K H800 GPU hours, or $294K at $2 per GPU hour, run on 64 x 8 H800 GPUs. Within that figure, R1-Zero accounts for $202K, creating the SFT data for $10K and R1 itself for $82K. The estimate excludes the cost of the DeepSeek-V3 base model. [4]
Main features
On the results DeepSeek reported against OpenAI-o1-1217, R1 scored 79.8 to 79.2 on AIME 2024, 97.3 to 96.4 on MATH-500, and 96.3 to 96.6 on Codeforces percentile, which corresponds to a rating of 2029 against 2061. [1]
The code and weights are under the MIT License, which explicitly allows commercial use and distillation. DeepSeek also released distilled models at 1.5B, 7B, 8B, 14B, 32B and 70B parameters, based on Qwen2.5 and Llama 3. [2]
The API launched at $0.55 per million input tokens on a cache miss, $0.14 on a cache hit and $2.19 per million output tokens, under the model name deepseek-reasoner. [7]
By 27 January 2025 the DeepSeek chatbot app had overtaken ChatGPT as the most downloaded free app on the U.S. iOS App Store. [5]
On 28 May 2025 DeepSeek updated the model as DeepSeek-R1-0528, which raised AIME 2025 accuracy from 70.0% to 87.5%. [8]
The paper was published in Nature on 17 September 2025 (volume 645, pages 633–638), with Liang Wenfeng as corresponding author. [3]
Its place in development
From DeepSeek-V3.1, released on 21 August 2025, the separate reasoner was folded into one hybrid model. The deepseek-reasoner API name was mapped to V3.1's thinking mode. [9]
CapabilitiesLarge language modelsReasoning
Ecosystem & regionOpen sourceChina
Version history
12-
2023
-
Nov 29, 2023
DeepSeek LLM (7B and 67B)
DeepSeek LLM is DeepSeek's first general-purpose language model series, released on 29 November 2023 in 7B and 67B sizes, each with a base version and a chat version.
-
Nov 29, 2023
DeepSeek LLM (7B and 67B)
-
2024
-
May 7, 2024
DeepSeek-V2
DeepSeek-V2 is a mixture-of-experts language model released in May 2024 with 236B total parameters, of which 21B are activated per token. It introduced Multi-head Latent Attention and used DeepSeekMoE for economical sparse training.
-
Sep 5, 2024
DeepSeek-V2.5
DeepSeek-V2.5 is a model released on 5 September 2024 that merged DeepSeek's chat and code models into one, keeping both conversational and coding ability while improving safety scores.
-
Dec 26, 2024
DeepSeek-V3
DeepSeek-V3 is an open-weight mixture-of-experts language model that DeepSeek released in December 2024. Each token uses 37 billion of its 671 billion parameters, and DeepSeek describes its chat model as comparable to leading closed-source systems.
-
May 7, 2024
DeepSeek-V2
-
2025
-
Jan 20, 2025
DeepSeek-R1 · This page
DeepSeek-R1 is an open-weight reasoning model that DeepSeek released on 20 January 2025. Built on DeepSeek-V3-Base and trained with reinforcement learning, it is released under the MIT License, which allows commercial use and distillation.
-
Mar 24, 2025
DeepSeek-V3-0324
DeepSeek-V3-0324 is a March 2025 update to DeepSeek-V3 that keeps the same model structure while raising benchmark scores and moving to the MIT License.
-
May 28, 2025
DeepSeek-R1-0528
An updated release of DeepSeek's open-weight reasoning model, published on 28 May 2025 under the MIT License, with markedly higher scores on competition math, science and coding benchmarks, plus system-prompt support and stronger function calling.
-
Aug 21, 2025
DeepSeek-V3.1
DeepSeek-V3.1, released on 21 August 2025 under the MIT License, was DeepSeek's first hybrid model, supporting both a thinking mode and a non-thinking mode in one set of weights. It has 671 billion total parameters, with 37 billion activated per token.
-
Sep 29, 2025
DeepSeek-V3.2-Exp
An experimental DeepSeek model released on 29 September 2025 that introduced DeepSeek Sparse Attention to make long-context training and inference cheaper, shipped with open weights and an immediate API price cut.
-
Dec 1, 2025
DeepSeek-V3.2 and V3.2-Speciale
DeepSeek-V3.2, released on 1 December 2025, succeeded the experimental V3.2-Exp and integrated thinking into tool use. Its companion V3.2-Speciale, which maximizes reasoning, reached gold-level results at several 2025 competitions.
-
Jan 20, 2025
DeepSeek-R1 · This page
-
2026
-
Apr 24, 2026
DeepSeek-V4 (V4-Pro and V4-Flash)
DeepSeek-V4, released in preview on 24 April 2026, comprises V4-Pro and V4-Flash, two open mixture-of-experts models with a 1M-token context. Its architecture targets long-context efficiency, and its later releases extended the series with agent upgrades and multimodal input.
-
Sep 10, 2026
DeepSeek-V4.1-Flash
DeepSeek-V4.1-Flash, released on 10 September 2026 under the API name deepseek-flash, is a 552-billion-parameter multimodal mixture-of-experts model with a causal encoder-decoder design and a 1M-token context.
-
Apr 24, 2026
DeepSeek-V4 (V4-Pro and V4-Flash)
20257
-
ReleaseMiniMax releases MiniMax-M1, an open reasoning model with 1M context
In June 2025 MiniMax released MiniMax-M1, an open reasoning model with a native 1-million-token context. It has 456B total parameters and was released under Apache 2.0, and its design was compared with DeepSeek R1 in the release figures.
-
ReleaseTencent officially releases the Hunyuan-T1 reasoning model
On 21 March 2025 Tencent launched Hunyuan-T1, a deep-reasoning model offered on Tencent Cloud and priced against DeepSeek-R1.
-
ReleaseQwen releases QwQ-32B, matching DeepSeek-R1 at 32B parameters
On 6 March 2025 the Qwen team released QwQ-32B, a 32-billion-parameter reasoning model that it said matches DeepSeek-R1, which has 671B parameters, on its benchmarks. It is available with open weights.
-
ReleaseTencent releases Hunyuan Turbo S, a fast-thinking model
On 27 February 2025 Tencent released Hunyuan Turbo S, saying it replies faster than DeepSeek-R1. The model was offered through a Tencent Cloud API, and its weights were not released.
-
UpdateTencent Yuanbao integrates DeepSeek-R1
On 13 February 2025 Tencent's Yuanbao app added DeepSeek-R1 beside Tencent's own Hunyuan models, and within days it rose to second place on China's App Store free chart.
-
MilestoneDeepSeek app tops the U.S. App Store
On 27 January 2025 DeepSeek's mobile app passed ChatGPT as the most downloaded free app on Apple's U.S. App Store. The same day Nvidia shares fell 17 percent, the biggest one-day loss in U.S. history for any company.
-
ReleaseDeepSeek releases DeepSeek-R1 as an open reasoning model
On 20 January 2025 DeepSeek released DeepSeek-R1 as an open reasoning model, claiming performance on par with OpenAI o1 on math, code and reasoning tasks. Its release coincided with a sharp drop in Nvidia's share price.
Developed by1
Builds on3
Succeeded by1
Powers2
Notes and sources
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learninghttps://arxiv.org/abs/2501.12948
- DeepSeek-R1 repositoryhttps://github.com/deepseek-ai/DeepSeek-R1
- DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning (Nature 645, 633–638)https://pmc.ncbi.nlm.nih.gov/articles/PMC12443585/
- Supplementary Information for DeepSeek-R1 (Nature)https://media.springernature.com/original/springer-static/esm/art%3A10.1038%2Fs41586-025-09422-z/MediaObjects/41586_2025_9422_MOESM1_ESM.pdf
- DeepSeekhttps://en.wikipedia.org/wiki/DeepSeek
- DeepSeek-R1-Lite Releasehttps://api-docs.deepseek.com/news/news1120
- DeepSeek-R1 Releasehttps://api-docs.deepseek.com/news/news250120
- deepseek-ai/DeepSeek-R1-0528https://huggingface.co/deepseek-ai/DeepSeek-R1-0528
- DeepSeek V3.1 Releasehttps://api-docs.deepseek.com/news/news250821
Cite this page: DeepSeek-R1. TellWorld, https://www.tellworld.net/models/deepseek-r1. Updated 2026-11-08