DeepSeek

DeepSeek

deepseek.comCN
+
+
+
+
Portfolio Stats
Total Models5
Multimodal0
Benchmarks Run5
Avg Performance27.3%
+
+
+
+
Latest Release
DeepSeek-V3.2
Released: Dec 1, 2025
+
+
+
+
Release Timeline
Recent model releases by year
2025
5 models
+
+
+
+
Performance Overview
Top models and benchmark performance

Top Performing Models

By avg score
#1DeepSeek-V3.2 Thinking
73.1%
#2DeepSeek-V3.2
59.0%
#3DeepSeek-R1
3.1%
#4DeepSeek-V3.1
1.3%
#5DeepSeek-V3.2-Speciale
0.0%

Benchmark Categories

Coding
2
55.3%
Agents
3
28.3%

Model Statistics

Multimodal Ratio
0%
Models with Providers
5

All Models

Complete portfolio of 5 models with advanced filtering

LicenseLinks
#01DeepSeekDeepSeek-V3.2 Thinking
DeepSeek-V3.2-Exp, the experimental thinking release of DeepSeek-V3.2 with 685B parameters
Sep 29, 2025
MIT
73.1%
#02DeepSeekDeepSeek-V3.2
DeepSeek-V3.2 with 685B parameters, agentic tool-use, thinking modes, and 128K context
Dec 1, 2025
MIT
-
#03DeepSeekDeepSeek-V3.2-Speciale
DeepSeek-V3.2-Speciale, a high-compute variant of V3.2 with gold-medal reasoning performance
Dec 1, 2025
MIT
-
#04DeepSeekDeepSeek-V3.1
DeepSeek-V3.1 with 671B parameters, hybrid thinking and non-thinking modes, and 128K context
Aug 21, 2025
MIT
-
#05DeepSeekDeepSeek-R1
DeepSeek's R1 reasoning model with 671B parameters, 128K context, and extended chain-of-thought
Jan 20, 2025
MIT
-
+
+
+
+
Resources
Official Website