Home/Models/DeepSeek V3.2
DeepSeek logo

DeepSeek V3.2

by DeepSeek · Released Dec 2025

Open Source
58.7
avg score
Rank #73
Compare
Better than 69% of all models
Context
131K tokens (~66 books)
Input $/1M
$0.25
Output $/1M
$0.38
Type
text
License
Open Source
Benchmarks
29 tested
Data updated today
About

DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...

Tested on 29 benchmarks with 53.0% average. Top scores: Chatbot Arena Elo — Overall (1424.4%), Chatbot Arena Elo — Coding (1326.9%), OpenCompass — AIME2025 (93.0%).

Capabilities
coding
62.3
#33 globally
reasoning
37.6
#71 globally
math
53.8
#67 globally
knowledge
52.1
#89 globally
language
59.0
#88 globally
speed
73.0
#20 globally
Benchmark Scores
Compare All
Tested on 29 benchmarks · Ranked across 7 categories
Score Distribution (all 233 models)
0255075100
▲ You are here
LiveBench — Coding

Regularly refreshed coding problems that avoid data contamination. New problems added monthly to prevent memorization.

75.7
OpenCompass — LiveCodeBenchV6

OpenCompass Live Code Bench v6. Fresh competitive programming problems to evaluate code generation without memorization.

75.4
Aider polyglot

Multi-language code editing from Aider. Tests editing ability across Python, JavaScript, TypeScript, Java, C++, Go, Rust, and more.

74.2
ARC-AGI

Abstraction and Reasoning Corpus. Tests fluid intelligence through novel visual pattern recognition puzzles. Core measure of general intelligence.

57.0
LiveBench — Data Analysis

Fresh data analysis tasks testing ability to interpret tables, charts, and statistical data.

45.0
LiveBench — Reasoning

Regularly refreshed reasoning problems testing logical deduction, spatial reasoning, and analytical thinking.

44.3
OpenCompass — AIME2025

OpenCompass evaluation on AIME 2025 problems. Tests mathematical reasoning on fresh competition problems.

93.0
OTIS Mock AIME 2024-2025

Mock AIME (American Invitational Mathematics Exam) problems from OTIS. Tests mathematical competition performance.

87.8
LiveBench — Mathematics

Regularly updated math problems that test numerical reasoning, algebra, calculus, and combinatorics.

64.0
Excellent (85+) Good (70-85) Average (50-70) Below (<50)
Recently Happened
DeepSeek V3.2 output price dropped · $0.38 per 1M tokens
Mar 28, 2026
Links
Documentation
Community
BenchGecko API
deepseek-v3-2
Specifications
  • Typetext
  • Context131K tokens (~66 books)
  • ReleasedDec 2025
  • LicenseOpen Source
  • StatusActive
  • Cost / Message~$0.001
Available On
DeepSeek logoDeepSeek$0.25
Share & Export
Tweet
DeepSeek V3.2 is an open-source text AI model by DeepSeek, released in December 2025. It has an average benchmark score of 58.7. Context window: 131K tokens.