DeepSeek V3.2 Exp (Non-reasoning)

Last updated: Recently

Model Type

Agent

Max Context

131K

Max Output

65,536 tokens

Parameter Scale

DeepSeek V3.2 is large language model designed to harmonize high computational efficiency with reasoning and agentic tool use performance. 131,072 token context window, maximum output of 65,536 tokens. Higher uptime with 9 providers. Includes independent benchmarks from Artificial Analysis.

Input Modality

text

Output Modality

text

Inference Speed

Latest Release Date

9/29/2025

SDK Ecosystem

artificial-analysismanufactureraa-bootstrap

Model Overview Fields

name

DeepSeek V3.2 Exp (Non-reasoning)

Release date

9/29/2025

Performance

model_creator.name

DeepSeek

AA Evaluation Scores

AA Intelligence Index

28.4

AA Coding Index

AA Math Index

57.7

MMLU Pro

0.836

GPQA

0.738

HLE

0.086

LiveCodeBench

0.554

SciCode

0.399

Math-500

AIME

AIME 2025

0.576666666666667

IFBench

0.431292517006803

LCR

0.43

TerminalBench Hard

0.25

TAU2

0.339181286549708

Output Speed (tokens/s)

TTFT (s)

First Answer Token (s)

DeepSeek

DeepSeek V3.2 Exp (Non-reasoning)

$0.28/M$0.41/M

There are no reviews for this model yet.

artificial-analysismanufactureraa-bootstrap

0.0