Skip to main content
leaderboard
.cn
Toggle Sidebar
Search models or organizations…
⌘K
Explore
Frontier Models
Historical Trends
Local Models
Compare Models
Ranking Studio
Library
Models
Pricing
Organizations
Benchmarks
Data sources
Settings
Toggle Sidebar
leaderboard
.cn
Loading page content
Feedback
Loading page content
Back to models
DeepSeek-V3.2
Open
Share
Add to comparison
Tool use · Reasoning
Overview
Deployment
Pricing
Evaluations
Evaluation overview
Overall rank
#70
70.4 points · 180 models
Open-weight rank
#21
78.4 points · 109 models
Data coverage
27 items
5 sources · updated through 2026-07-16
Domain performance
Coding
#63 / 160
72.5
Research
#67 / 174
81.2
Agents
#58 / 125
68.7
Evaluation highlights
All evaluations
AIME 2025
Reasoning · GLM-4.7 Technical Report
93.1%
9
/ 60
Terminal Bench Hard
Coding · GLM-4.7 Technical Report
35.4%
3
/ 8
τ²-Bench
Agents · GLM-5 Technical Report
85.3%
8
/ 21
Arena Text Rating
Reasoning · Arena.ai Text
1,425.1Elo
71
/ 354
MMLU-Pro
Reasoning · GLM-4.7 Technical Report
85%
28
/ 95
GPQA-Diamond
Reasoning · GLM-5.1 Technical Report
82.4%
53
/ 126
Basic information
Context (input / output)
164K / 16K
Parameters
671B / 37B active
Release date
2025-12
License
MIT
Input modalities
Text
Output modalities
Text
Active parameters
37B
Availability
DeepSeek
Full pricing
Input
≈ $0.2857 /1M
Output
≈ $1.1429 /1M
vLLM · FP8
Deployment options
FP8 · 8 × 141 GB VRAM
Related links
Official site
HuggingFace
Version relationships
DeepSeek-V3 family
DeepSeek-V3
2024-12 · Same series
DeepSeek-V3 Base
2024-12 · Same series
DeepSeek-V3-0324
2025-03 · Same series
DeepSeek-V3.2
Current
2025-12 · Current model
Runtime modes
Thinking