DeepSeek-R1 logo

DeepSeek-R1

DeepSeek

DeepSeek-R1 is a cutting-edge reasoning model developed using DeepSeek-V3 as its foundation (671B total parameters, 37B activated per token). This first-generation model leverages extensive reinforcement learning (RL) to significantly improve its chain-of-thought processes and overall reasoning abilities. As a result, DeepSeek-R1 excels in complex tasks involving mathematics, coding, and multi-step reasoning.

Model Specifications

Technical details and capabilities of DeepSeek-R1

Core Specifications

671.0B Parameters

Model size and complexity

14800.0B Training Tokens

Amount of data used in training

131.1K / 131.1K

Input / Output tokens

January 19, 2025

Release date

Capabilities & License

Multimodal Support
Not Supported
Web Hydrated
No
License
MIT License

Resources

Research Paper
https://arxiv.org/abs/2501.12948
API Reference
https://api-docs.deepseek.com/news/news250120
Playground
https://chat.deepseek.com
Code Repository
https://github.com/deepseek-ai/DeepSeek-R1

Performance Insights

Check out how DeepSeek-R1 handles various AI tasks through comprehensive benchmark results.

100
75
50
25
0
97.3
MATH-500
97.3
(97%)
92.9
MMLU-Redux
92.9
(93%)
92.8
CLUEWSC
92.8
(93%)
92.3
ArenaHard
92.3
(92%)
92.2
DROP
92.2
(92%)
91.8
C-Eval
91.8
(92%)
90.8
MMLU
90.8
(91%)
87.6
AlpacaEval2.0
87.6
(88%)
84
MMLU-Pro
84
(84%)
83.3
IFEval
83.3
(83%)
82.5
FRAMES
82.5
(83%)
79.8
AIME 2024
79.8
(80%)
78.8
CNMO 2024
78.8
(79%)
71.5
GPQA
71.5
(72%)
70
AIME 2025
70
(70%)
65.9
LiveCodeBench
65.9
(66%)
63.7
C-SimpleQA
63.7
(64%)
56.9
Aider Polyglot
56.9
(57%)
53.3
Aider-Polyglot
53.3
(53%)
49.2
SWE-bench Verified
49.2
(49%)
30.1
SimpleQA
30.1
(30%)
8.6
Humanity's Last Exam
8.6
(9%)
MATH-500
MMLU-Redux
CLUEWSC
ArenaHard
DROP
C-Eval
MMLU
AlpacaEval2.0
MMLU-Pro
IFEval
FRAMES
AIME 2024
CNMO 2024
GPQA
AIME 2025
LiveCodeBench
C-SimpleQA
Aider Polyglot
Aider-Polyglot
SWE-bench Verified
SimpleQA
Humanity's Last Exam

Model Comparison

See how DeepSeek-R1 stacks up against other leading models across key performance metrics.

100
80
60
40
20
0
90.8
MMLU - DeepSeek-R1
90.8
(91%)
88.5
MMLU - DeepSeek-V3
88.5
(89%)
88.7
MMLU - GPT-4o
88.7
(89%)
87.3
MMLU - Llama 3.1 405B Instruct
87.3
(87%)
84.8
MMLU - Phi-4
84.8
(85%)
83.6
MMLU - Llama 3.1 70B Instruct
83.6
(84%)
84
MMLU-Pro - DeepSeek-R1
84
(84%)
75.9
MMLU-Pro - DeepSeek-V3
75.9
(76%)
72.6
MMLU-Pro - GPT-4o
72.6
(73%)
73.3
MMLU-Pro - Llama 3.1 405B Instruct
73.3
(73%)
70.4
MMLU-Pro - Phi-4
70.4
(70%)
66.4
MMLU-Pro - Llama 3.1 70B Instruct
66.4
(66%)
92.2
DROP - DeepSeek-R1
92.2
(92%)
91.6
DROP - DeepSeek-V3
91.6
(92%)
83.4
DROP - GPT-4o
83.4
(83%)
84.8
DROP - Llama 3.1 405B Instruct
84.8
(85%)
75.5
DROP - Phi-4
75.5
(76%)
79.6
DROP - Llama 3.1 70B Instruct
79.6
(80%)
83.3
IFEval - DeepSeek-R1
83.3
(83%)
86.1
IFEval - DeepSeek-V3
86.1
(86%)
84
IFEval - GPT-4o
84
(84%)
88.6
IFEval - Llama 3.1 405B Instruct
88.6
(89%)
63
IFEval - Phi-4
63
(63%)
87.5
IFEval - Llama 3.1 70B Instruct
87.5
(88%)
71.5
GPQA - DeepSeek-R1
71.5
(72%)
59.1
GPQA - DeepSeek-V3
59.1
(59%)
53.6
GPQA - GPT-4o
53.6
(54%)
50.7
GPQA - Llama 3.1 405B Instruct
50.7
(51%)
56.1
GPQA - Phi-4
56.1
(56%)
41.7
GPQA - Llama 3.1 70B Instruct
41.7
(42%)
MMLU
MMLU-Pro
DROP
IFEval
GPQA
DeepSeek-R1
DeepSeek-V3
GPT-4o
Llama 3.1 405B Instruct
Phi-4
Llama 3.1 70B Instruct

Detailed Benchmarks

Dive deeper into DeepSeek-R1's performance across specific task categories. Expand each section to see detailed metrics and comparisons.

Math

AIME 2024

o3
96.7%
o3-mini
87.3%
o1-pro
86.0%
o1
83.3%
Claude 3.7 Sonnet
Claude 3.7 Sonnet
80.0%
DeepSeek-R1
DeepSeek-R1
79.8%
Kimi-k1.5
77.5%
o1-preview
42.0%
GPT-4.5
36.7%
GPT-4o
13.4%
Current model
Other models
Avg (68.3%)

MATH-500

DeepSeek-R1
DeepSeek-R1
97.3%
Claude 3.7 Sonnet
Claude 3.7 Sonnet
96.2%
Kimi-k1.5
96.2%
QwQ-32B-Preview
90.6%
DeepSeek-V3
DeepSeek-V3
90.2%
o1-mini
90.0%
Current model
Other models
Avg (93.4%)

AIME 2025

Grok-3
Grok-3
93.0%
Grok-3 Mini
Grok-3 Mini
90.3%
Gemini Pro 2.5 Experimental
Gemini Pro 2.5 Experimental
86.7%
o3-mini
86.5%
DeepSeek-R1
DeepSeek-R1
70.0%
Claude 3.7 Sonnet
Claude 3.7 Sonnet
49.5%
Current model
Other models
Avg (79.3%)

Coding

LiveCodeBench

Grok-3 Mini
Grok-3 Mini
80.0%
Grok-3
Grok-3
79.0%
o3-mini
74.1%
Gemini Pro 2.5 Experimental
Gemini Pro 2.5 Experimental
70.4%
DeepSeek-R1
DeepSeek-R1
65.9%
QwQ 32B
QwQ 32B
63.4%
Kimi-k1.5
62.5%
Qwen2.5 72B Instruct
Qwen2.5 72B Instruct
55.5%
Qwen2.5-Coder 7B Instruct
Qwen2.5-Coder 7B Instruct
18.2%
Current model
Other models
Avg (63.2%)

SWE-bench Verified

Claude 3.7 Sonnet
Claude 3.7 Sonnet
70.3%
Gemini Pro 2.5 Experimental
Gemini Pro 2.5 Experimental
63.8%
o3-mini
49.3%
DeepSeek-R1
DeepSeek-R1
49.2%
Claude 3.5 Sonnet
Claude 3.5 Sonnet
49.0%
o1
48.9%
DeepSeek-V3
DeepSeek-V3
42.0%
Claude 3.5 Haiku
Claude 3.5 Haiku
40.6%
GPT-4.5
38.0%
Current model
Other models
Avg (50.1%)

Aider-Polyglot

DeepSeek-R1
DeepSeek-R1
53.3%
DeepSeek-V3
DeepSeek-V3
49.6%
Current model
Other models
Avg (51.4%)

Aider Polyglot

Gemini Pro 2.5 Experimental
Gemini Pro 2.5 Experimental
74.0%
Gemini Pro 2.5 Experimental
Gemini Pro 2.5 Experimental
72.9%
Claude 3.7 Sonnet
Claude 3.7 Sonnet
64.9%
o1
61.7%
o3-mini
60.4%
DeepSeek-R1
DeepSeek-R1
56.9%
GPT-4.5
44.9%
GPT-4o
27.1%
QwQ 32B
QwQ 32B
20.9%
Current model
Other models
Avg (53.7%)

Reasoning

DROP

DeepSeek-R1
DeepSeek-R1
92.2%
DeepSeek-V3
DeepSeek-V3
91.6%
Claude 3.5 Sonnet
Claude 3.5 Sonnet
87.1%
Claude 3.5 Sonnet
Claude 3.5 Sonnet
87.1%
GPT-4 Turbo
86.0%
Nova Pro
85.4%
Llama 3.1 405B Instruct
Llama 3.1 405B Instruct
84.8%
GPT-4o
83.4%
Current model
Other models
Avg (87.2%)

Knowledge

MMLU

o1
91.8%
DeepSeek-R1
DeepSeek-R1
90.8%
o1-preview
90.8%
Claude 3.5 Sonnet
Claude 3.5 Sonnet
90.4%
Claude 3.5 Sonnet
Claude 3.5 Sonnet
90.4%
GPT-4o
88.7%
DeepSeek-V3
DeepSeek-V3
88.5%
GPT-4o
88.0%
Current model
Other models
Avg (89.9%)

GPQA

o3
87.7%
o1-pro
79.0%
o1
78.0%
Gemini 2.0 Flash Thinking
Gemini 2.0 Flash Thinking
74.2%
o1-preview
73.3%
DeepSeek-R1
DeepSeek-R1
71.5%
GPT-4.5
71.4%
Claude 3.5 Sonnet
Claude 3.5 Sonnet
67.2%
QwQ-32B-Preview
65.2%
Qwen2 7B Instruct
Qwen2 7B Instruct
25.3%
Current model
Other models
Avg (69.3%)

Non categorized

CLUEWSC

DeepSeek-R1
DeepSeek-R1
92.8%
Kimi-k1.5
91.4%
DeepSeek-V3
DeepSeek-V3
90.9%
Current model
Other models
Avg (91.7%)

MMLU-Redux

DeepSeek-R1
DeepSeek-R1
92.9%
DeepSeek-V3
DeepSeek-V3
89.1%
Qwen2.5 72B Instruct
Qwen2.5 72B Instruct
86.8%
Qwen2.5 32B Instruct
Qwen2.5 32B Instruct
83.9%
Qwen2.5 14B Instruct
Qwen2.5 14B Instruct
80.0%
Qwen2.5-Coder 32B Instruct
Qwen2.5-Coder 32B Instruct
77.5%
Qwen2.5 7B Instruct
Qwen2.5 7B Instruct
75.4%
Qwen2.5-Coder 7B Instruct
Qwen2.5-Coder 7B Instruct
66.6%
Current model
Other models
Avg (81.5%)

MMLU-Pro

DeepSeek-R1
DeepSeek-R1
84.0%
Claude 3.5 Sonnet
Claude 3.5 Sonnet
77.6%
Gemini 2.0 Flash
Gemini 2.0 Flash
76.4%
Claude 3.5 Sonnet
Claude 3.5 Sonnet
76.1%
DeepSeek-V3
DeepSeek-V3
75.9%
Gemini 1.5 Pro
Gemini 1.5 Pro
75.8%
Grok-2
Grok-2
75.5%
GPT-4o
74.7%
Current model
Other models
Avg (77.0%)

IFEval

Claude 3.7 Sonnet
Claude 3.7 Sonnet
93.2%
Olmo 2 32B
Olmo 2 32B
85.6%
Qwen2.5 72B Instruct
Qwen2.5 72B Instruct
84.1%
GPT-4o
84.0%
QwQ 32B
QwQ 32B
83.9%
DeepSeek-R1
DeepSeek-R1
83.3%
Mistral Small 3
Mistral Small 3
82.9%
Llama 3.1 8B Instruct
Llama 3.1 8B Instruct
80.4%
Llama 3.2 3B Instruct
Llama 3.2 3B Instruct
77.4%
Pixtral-12B
61.3%
Current model
Other models
Avg (81.6%)

SimpleQA

GPT-4.5
62.5%
Gemini Pro 2.5 Experimental
Gemini Pro 2.5 Experimental
52.9%
Grok-3
Grok-3
43.6%
o1
42.6%
o1-preview
42.4%
DeepSeek-R1
DeepSeek-R1
30.1%
DeepSeek-V3
DeepSeek-V3
24.9%
o3-mini
13.8%
Mistral Small 3.1 24B
Mistral Small 3.1 24B
10.4%
Phi-4
3.0%
Current model
Other models
Avg (32.6%)

FRAMES

DeepSeek-R1
DeepSeek-R1
82.5%
DeepSeek-V3
DeepSeek-V3
73.3%
Current model
Other models
Avg (77.9%)

ArenaHard

DeepSeek-R1
DeepSeek-R1
92.3%
DeepSeek-V2.5
DeepSeek-V2.5
76.2%
Phi-4
75.4%
Current model
Other models
Avg (81.3%)

C-Eval

DeepSeek-R1
DeepSeek-R1
91.8%
DeepSeek-V3
DeepSeek-V3
86.5%
Qwen2 72B Instruct
Qwen2 72B Instruct
83.8%
Qwen2 7B Instruct
Qwen2 7B Instruct
77.2%
Current model
Other models
Avg (84.8%)

C-SimpleQA

DeepSeek-V3
DeepSeek-V3
64.8%
DeepSeek-R1
DeepSeek-R1
63.7%
Current model
Other models
Avg (64.3%)

Humanity's Last Exam

Gemini Pro 2.5 Experimental
Gemini Pro 2.5 Experimental
18.8%
o3-mini
14.0%
Claude 3.7 Sonnet
Claude 3.7 Sonnet
8.9%
DeepSeek-R1
DeepSeek-R1
8.6%
GPT-4.5
6.4%
Current model
Other models
Avg (11.3%)

Providers Pricing Coming Soon

We're working on gathering comprehensive pricing data from all major providers for DeepSeek-R1. Compare costs across platforms to find the best pricing for your use case.

OpenAI
Anthropic
Google
Mistral AI
Cohere

Share your feedback

Hi, I'm Charlie Palars, the founder of Deepranking.ai. I'm always looking for ways to improve the site and make it more useful for you. You can write me through this form or directly through X at @palarsio.

Your feedback helps us improve our service

Stay Ahead with AI Updates

Get insights on Gemini Pro 2.5, Sonnet 3.7 and more top AI models