NousCoder-14B

NousCoder-14B

Writes code for competitive programming problems, a Qwen3-14B model trained further with reinforcement learning on 24k problems with checkable answers. For developers who run a coding model on their own computer.

Made by NousResearch on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: NousCoder-14B is a text generation model from NousResearch for competitive programming. It starts from Qwen3-14B and was trained further with reinforcement learning on 24k coding problems whose answers can be checked automatically. It has 14.8B parameters and ships as Safetensors files.

Who it suits: It suits people who want a coding model they can run themselves, especially for contest style problems with a clear right or wrong answer.

Industry | Who uses it | How they use it | Example
Education | Programming students | Practice contest style problems and compare their own solution with the model's | An algorithm exercise with a clear pass or fail result
Software | Developers | Run a code model on their own hardware to draft solutions | A function drafted from a written problem statement
Research | Machine learning researchers | Study a model trained with reinforcement learning on checkable problems | Comparing it with its Qwen3-14B base on coding tasks
Contests | Competitive programmers | Get a second approach to a problem while practising | A different solution to a practice problem

You can skip it if you do not write code or lack the memory to run a 14.8B model.

Where to use it:

• Your own computer, on Windows, macOS or Linux with Python, to download the files
• Any local tool that loads Safetensors text models
• Quantized copies made by other people, for machines with less memory

Use case | Who | What they get back
Solve a practice problem | Students | Program code for the problem you describe
Draft a solution | Developers | A first version of a function to review and test
Compare coding models | Researchers | A model to test against its Qwen3-14B base

How to set it up:

• Python and pip
• Memory for the weights: about 29.5 GB in BF16, or about 7.4 GB at 4-bit by estimate

1. Install the Hugging Face hub package.
2. Download the model files to your computer.
3. Open them in the local tool you use to run text models.

pip install -U huggingface_hub
hf download NousResearch/NousCoder-14B

How to use it: The model card gives no run command of its own. Once the files load in your tool, describe a programming problem in plain text and read the code it writes back. Run that code against your own test cases before you trust it.

Example: Ask it to write a program that reads a list of integers and prints the length of the longest increasing run. It replies with a program that tries to solve that task, which you then test.

For newcomers and experts: Think of this as a model that turns a written coding problem into code. For a first try, download the files with the two commands above, load them in a tool you already use for local models, and give it one small problem.

Underneath, it is Qwen3-14B after post-training with reinforcement learning. Training ran for four days on 48 B200 GPUs, with 24k coding problems that can be checked. Its reported result on LiveCodeBench v6 is a Pass@1 of 67.87 percent, against 60.79 percent for the base model.

Coding problems with checkable answers
Qwen3-14B is the starting point
Reinforcement learning on 24k problems
NousCoder-14B is the result
You send a problem, it writes code

Good to know: The score below is the author's own report on one benchmark, so test the model on your own problems.

Title: Pass@1 on LiveCodeBench v6
NousCoder-14B: 67.87 percent
Qwen3-14B: 60.79 percent

• License: Apache 2.0
• Cost to run: no API key or paid service is named, you supply the hardware
• Data it reaches: downloads model files from Hugging Face, and no other access is named
• Not for: the card describes competitive programming only

On Hugging Face, 13 November 2026

• Task: text generation
• Parameters: 14.8B
• Weights: about 29.5 GB in BF16; about 7.4 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 839
• Likes: 233
• License: apache-2.0
• Based on: Qwen/Qwen3-14B
• Quantized versions: 20
• Fine-tunes: 3
• Tags: safetensors, qwen3, text-generation, conversational

Links

• Model: https://huggingface.co/NousResearch/NousCoder-14B
• Files: https://huggingface.co/NousResearch/NousCoder-14B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download NousResearch/NousCoder-14B

Then follow the usage in its model card. Nothing in this kit runs by itself.