DeepSeek Reasoner

DeepSeek-R1 is now live and open source, rivaling OpenAI's Model o1.

Pricing

Auto routing · per 1M tokens

Input: $0.40
Output: $1.70

Compare provider prices

Specifications

Context window: 64K
Max output: 65.5K
Parameters: 671B / 37B; Total / active
Avg output (7d): 708 tokens; Longer than 57% of models

Benchmarks

Artificial Analysis

LMArena

Vectara

Sourced from Artificial Analysis.

Intelligence Index

13.1

Better than 51% of models compared

Coding Index

24.6

Better than 30% of models compared

Agentic Index

1.1

Better than 19% of models compared

Agentic work

T²-Bench Telecom (legacy)

Legacy fallback · Conversational AI agents in dual-control scenarios

36.5%

Better than 44% of models compared

GDPval-AA (unversioned / legacy)

Legacy fallback · Economically valuable tasks

0.0%

Document reasoning

AA-LCR v1.1

Long context reasoning with updated grading

55.7%

Better than 48% of models compared

Reasoning

HLE

Humanity's Last Exam

15.8%

Better than 63% of models compared

IFBench

Instruction-following benchmark

39.6%

Better than 38% of models compared

CritPt

Research-level physics reasoning

0.6%

Coding

Terminal-Bench Hard (legacy)

Legacy fallback · Agentic coding and terminal use

15.9%

Better than 53% of models compared

SciCode

Python programming for scientific computing

38.3%

Better than 20% of models compared

LiveCodeBench

Contamination-free coding benchmark

77.0%

Better than 88% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

76.0%

Better than 71% of models compared

AIME

American Invitational Mathematics Examination

89.3%

Better than 95% of models compared

Math-500

Diverse mathematical problem solving benchmark

98.3%

Better than 94% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

84.9%

Better than 90% of models compared

AA-Omniscience Accuracy

Proportion of correctly answered questions

30.5%

AA-Omniscience Hallucination Rate

Rate of incorrect answers among non-correct responses

90.3%

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

81.3%

Better than 69% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation