ModelBeat
Models/DeepSeek-V3.2
All models
D

DeepSeek: DeepSeek-V3.2

deepseek/deepseek-v3.2

DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance.

Context
131K
Input
$0.23 / 1M
Output
$0.34 / 1M
Released
Dec 1, 2025
Modalities
texttext

Benchmarks

Scores on standardized evaluations. Higher is better — and rank shows where DeepSeek-V3.2 lands among all models tracked on Model Beat.

30.0
Intelligence Index
Epoch AI
30th percentile of tracked models
27.0
Coding Index
Epoch AI
27th percentile of tracked models
29.0
Agentic Index
Epoch AI
29th percentile of tracked models

Reasoning

6 evals
ARC-AGI57.0%

Abstract visual reasoning

ARC-AGI-24.0%

Harder abstract reasoning

GPQA Diamond83.4%

Graduate-level scientific reasoning

SimpleBench52.6%

Common-sense trick questions

SimpleQA Verified27.5%

Factual accuracy & hallucination

WeirdML46.7%

Novel ML problem-solving

Coding

1 evals
WebDev Arena1286

Human-rated web development

Math

3 evals
AIME 2024/202587.8%

Olympiad-qualifier math

FrontierMath22.1%

Research-level math problems

FrontierMath Tier 42.1%

Hardest research math

Agentic & Tools

2 evals
APEX7.0%

Multi-step agentic tasks

Terminal-Bench39.6%

Command-line agentic tasks

In the news

1

Model & benchmark data from Epoch AI (CC BY); pricing, specs & descriptions from OpenRouter.