DeepSeek V3.1 (Reasoning)

Last updated: Recently

Model Type

Text

Max Context

164K

Max Output

32,768 tokens

Parameter Scale

DeepSeek V3.1 is large hybrid reasoning model 671B parameters, 37B active that supports both thinking and non thinking modes via prompt templates. 163,840 token context window, maximum output of 32,768 tokens. Higher uptime with 7 providers. Includes independent benchmarks from Artificial Analysis.

Input Modality

text

Output Modality

text

Inference Speed

Latest Release Date

8/21/2025

SDK Ecosystem

artificial-analysismanufactureraa-bootstrap

Model Overview Fields

name

DeepSeek V3.1 (Reasoning)

Release date

8/21/2025

Performance

model_creator.name

DeepSeek

AA Evaluation Scores

AA Intelligence Index

27.7

AA Coding Index

29.7

AA Math Index

89.7

MMLU Pro

0.851

GPQA

0.779

HLE

0.13

LiveCodeBench

0.784

SciCode

0.391

Math-500

AIME

AIME 2025

0.896666666666667

IFBench

0.414965986394558

LCR

0.533333333333333

TerminalBench Hard

0.25

TAU2

0.374269005847953

Output Speed (tokens/s)

TTFT (s)

First Answer Token (s)

DeepSeek

DeepSeek V3.1 (Reasoning)

$0.59/M$1.69/M

There are no reviews for this model yet.

artificial-analysismanufactureraa-bootstrap

0.0