iilib
Model•active

DeepSeek-V3-0324

DeepSeek-V3-0324 is an open-weight 671B-parameter mixture-of-experts model with about 37B activated parameters and a 128K context window.

Official siteVerified Oct 4, 2026

Overview

This updated DeepSeek V3 checkpoint is intended for general text and coding workloads. Its official model card documents local serving, function calling, JSON output, and FIM completion.

Model specifications

Family
DeepSeek V3
Context
128,000 tokens
Input
Text
Output
Text
Open weights
Yes
Capabilities
Function Calling, Structured Output

Related entities