DeepSeek
DeepSeek

DeepSeek V3.2

2025-12-01

DeepSeek V3.2 is a large-scale Mixture-of-Experts language model that harmonizes high computational efficiency with frontier-level reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained mechanism that reduces attention complexity from quadratic to linear, significantly cutting training and inference costs in long-context scenarios. Through scalable reinforcement learning post-training, it achieves performance comparable to GPT-5, with gold-medal results on the 2025 International Mathematical Olympiad and Olympiad in Informatics. The model also features a large-scale agentic task synthesis pipeline that improves instruction following and tool use in complex interactive environments.

API|Reasoning|Open ModelMIT
Knowledge Cutoff
2025-03
Input → Output Format
Context Memory
164KIN164KOUT
Cost/1M Words
$0.252IN$0.378OUT
Calculate Cost

AI Performance Evaluation

Arena Overall Score
1424
±4
As of 2026-04-23
Overall Rank
No.64
44,738 Votes
Arena by Ability
Hard Prompts
1447±5No.63
Expert Knowledge
1447±12No.64
Instruction Following
1419±6No.55
Conversation Memory
1427±8No.64
Creative
1399±8No.59
Coding
1468±7No.65
Math
1428±11No.57
Arena by Occupation
Creative Writing
1410±7No.56
Social Sciences
1448±8No.60
Media
1395±7No.67
Business
1420±7No.66
Healthcare
1441±12No.73
Legal
1431±11No.70
Software
1456±6No.67
Mathematics
1438±14No.56
Overall
AA Intelligence Index
42%↑3%
LiveBench
50%↓11%
Reasoning & Math
AA Math Index
92%↑19%
GPQA Diamond
84%↑3%
HLE
22%↑5%
MMLU-Pro
86%↑4%
AIME 2025
92%↑18%
LB Reasoning
44%↓15%
LB Math
64%↓10%
LB Data
45%↓5%
Coding
AA Coding Index
37%↑3%
LiveCodeBench
86%↑21%
LB Coding
76%↑2%
LB Agentic
47%↑3%
TAU2
91%↑17%
TerminalBench
36%↑5%
SciCode
39%↓2%
Language & Instructions
IFBench
61%↑4%
AA-LCR
65%↑3%
Hallucination (HHEM)
6.3%↓4%
Factual (HHEM)
94%↑4%
LB Language
64%↓8%
LB IF
23%↓23%
Output Speed
Standard Mode
47tok/s↓35
First Output 1.26s
Reasoning Mode
77tok/s↓11
First Output 26.78s