News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
DeepSeek Coder

DeepSeek Coder

Open-source code generation models from DeepSeek

DeepSeek Coder is a series of open-source language models designed for software development tasks. Released in early 2024 by DeepSeek, these models range from 1.3 billion to 33 billion parameters. They are trained on two trillion tokens of code and natural language to support code completion and infilling.

Share on WhatsApp

  • Trained on 2 trillion tokens of code and text
  • Supports a 16K token context window
  • Available in sizes from 1.3B to 33B parameters
  • Permissive license allows commercial use
  • Trained on 87% code and 13% natural language

What it is

DeepSeek Coder comprises a family of language models built specifically for programming tasks. The series includes variants with parameter counts of 1.3 billion, 5.7 billion, 6.7 billion, and 33 billion. These models were created to address the limitations of closed-source systems by providing open access to high-performance code intelligence. They handle tasks such as code generation, completion, and infilling within a project context.

How it came about

The models were introduced in January 2024 through a research paper submitted to arXiv. The development team aimed to create open-source alternatives that could rival proprietary systems like Codex and GPT-3.5. The initial release included multiple model sizes to accommodate different computational resources and user needs, ranging from small on-device versions to larger server-grade instances.

Inside the model

Each model was trained from scratch on a dataset of two trillion tokens. This corpus consists of 87 per cent code and 13 per cent natural language in English and Chinese. The architecture uses a 16K token window to process large contexts. A fill-in-the-blank training task helps the model understand code structure for better infilling capabilities. The 33 billion parameter instruct version underwent further fine-tuning on two billion tokens of instruction data.

What it can do

Evaluations show strong performance on benchmarks such as HumanEval, MultiPL-E, MBPP, DS-1000, and APPS. The models achieve state-of-the-art results among publicly available code systems. They support multiple programming languages and handle project-level code completion effectively. The system surpasses several existing closed-source models in these specific coding tests.

How to use it

The models are available under a permissive license that allows both research and unrestricted commercial use. Weights can be downloaded from repositories like Hugging Face. The code repository itself uses the MIT License. Users can interact with the models via local deployment using libraries such as Transformers or through provided chat interfaces.

Questions people ask about DeepSeek Coder

What sizes of DeepSeek Coder are available?

The series includes models with 1.3 billion, 5.7 billion, 6.7 billion, and 33 billion parameters. This range allows users to select a model that fits their specific hardware constraints and performance requirements.

Can I use DeepSeek Coder for commercial projects?

Yes. The models are released under a permissive license that explicitly permits unrestricted commercial use. This makes them suitable for integration into private software products and enterprise workflows without additional fees.

What languages does DeepSeek Coder support?

The training data includes code and natural language in both English and Chinese. It supports multiple programming languages for code generation and completion tasks, leveraging a corpus that is 87 per cent code.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on DeepSeek Coder yet, so the facts on this page come from the sources listed above: arxiv.org/abs/2401.14196, huggingface.co/deepseek-ai/deepseek-coder-33b-instruct. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.