#36
DeepSeek V3
deepseek-chat-v3

DeepSeek V3 (0324 Release) is the foundational model for the V3 series, featuring a Mixture-of-Experts (MoE) architecture with 671B parameters (37B active). While a major leap from V2 with strong coding and reasoning abilities for its time, it lacks the refined hybrid inference and efficiency optimizations found in the later V3.1 and V3.2 versions.

Performance Metrics

Helpfulness
Instruction Following
Comprehension
Empathy
Creative Writing
74.25
Helpfulness
78
Empathy
72
Instruction Following
75
Creative Writing
72.25
Comprehension
74
Speed
Avg 85 tok/s

Model Specifications

Release Date
March 25, 2025
Lab
DeepSeek
Type
Open Source
Context Size
128K
Max Output Tokens
4.1K
Cost per 1M tokens
$0.27 / $1.10
Model Inputs
Text
Model Outputs
Text
Tool Calling
Enabled

Compare With Similar Models