DeepSeek V3.2

Reasoning

Creator

DeepSeek

Released

2025-12-01

Intelligence

41.7

Artificial Analysis Index

Coding

36.7

Artificial Analysis Index

In $/1M

$0.30

input tokens

Out $/1M

$0.45

output tokens

Blended $/1M

$0.34

3:1 blended

Speed

0

tokens / sec

Benchmark breakdown

Independent evaluation scores, normalized to 0–100.

GPQA Diamond
84
Humanity’s Last Exam
22
MMLU-Pro
86
SciCode
39
LiveCodeBench
86
MATH-500
AIME 2025
92
τ²-Bench (agentic)
91
Terminal-Bench Hard
36
IFBench
61
Via Artificial Analysis