DeepSeek · Code Models

DeepSeek Coder

Name: DeepSeek Coder
Author: DeepSeek

DeepSeek's first code models (1.3B to 33B), trained from scratch on 2 trillion tokens of mostly code, with project-level completion and infilling.

Official page

Run it locally

With Ollama installed, download and start the default version with:

ollama run deepseek-coder:6.7b

Variant	Download
deepseek-coder:6.7b	3.8GB
deepseek-coder:1.3b	776MB
deepseek-coder:33b	19GB

Memory: About 3.8GB for the default deepseek-coder:6.7b version (4-bit)
Formats: Safetensors (Hugging Face), GGUF (Ollama, llama.cpp, LM Studio)

The download size is roughly the memory (RAM or GPU VRAM) needed to load a variant; allow extra for long contexts. See our hardware guide.

Overview

DeepSeek Coder was trained on a corpus of 87% code and 13% natural language, using repository-level data to understand cross-file context.

Highlights

1.3B, 6.7B and 33B sizes
2 trillion training tokens
16K-token context window
Fill-in-the-middle support

Typical uses

Local code completion and assistant use on modest hardware.

Things to consider

Superseded by DeepSeek-Coder-V2 and newer coder models.
Download sizes refer to the default quantized versions in the Ollama library; full-precision weights are larger.

Specifications are taken from the model card on Hugging Face, the Ollama library and the provider’s announcement. Benchmark results depend on evaluation settings; see how AI benchmarks work.

#Code Models #Open Weights #Local AI #DeepSeek