K2-Horizon-7B

K2-Horizon-7B

Writes text, code and reasoning answers with a 512K-token context window, for developers and researchers who want an open dense model whose training data, code and checkpoints are all public.

Made by IFM 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: K2-Horizon-7B is the medium dense model in the K2-Horizon family from IFM. It is a decoder-only text generation model with a 7B core, and Hugging Face lists 9B parameters in total. Its context window reaches 524,288 tokens.

Who it suits: It suits developers and researchers who want an open dense model for agentic, coding, long-context and reasoning work, with training material they can inspect.

Title: Who uses this model
Industry | Who uses it | How they use it | Example
Software | Developers | Add coding and agent features backed by a local model | Ask it to fix a bug in a long file
Research | Machine learning researchers | Study how abilities change during training | Compare the intermediate checkpoints with the final model
AI labs | Training teams | Reuse the public data and recipe | Reproduce a training run with the released code

If you only need a tiny model for short chats, you can skip it, since this one has 9B parameters.

Where to use it:

• Python scripts that use the transformers library
• A GPU machine with room for the weights
• Windows, macOS or Linux with Python, for downloading the files

Title: Jobs it does
Use case | Who | What they get back
Coding help | Developers | Code and explanations in text
Long documents | Analysts | Answers drawn from input up to 524,288 tokens
Reasoning | Researchers | Written step-by-step answers
Training study | Researchers | Intermediate checkpoints to compare across training

How to set it up: You need Python, and disk space for the weights.

• Python with pip
• About 18 GB of disk space for BF16 weights, or about 4.5 GB for a 4-bit copy by Hugging Face's estimate
• The transformers library

1. Install huggingface_hub.
2. Run the download command to save the model files on your computer.
3. Load the files with transformers in your own script.

pip install -U huggingface_hub
hf download IFM/K2-Horizon-7B

How to use it: Load the downloaded files with transformers and prompt the model in English. The card also points to a separate repository, K2-Horizon-7B-Uno, which holds Diffusion Adapters for faster inference.

Example: You give the model a long codebase file and ask it to find the cause of a failing test. It replies in text with an explanation and a suggested fix.

For newcomers and experts: Think of it as a general text and code model that can read long inputs. A first small test is a short coding question. Move to long documents once that works.

It is a dense decoder-only model, not a mixture of experts. The long context is native from the midtraining stages onward, so it is not stretched afterward. The card compares it with Gemma 4-12B, Qwen3.5-9B and Granite 4.2-8B across math, coding and other benchmarks.

Text in
Dense decoder-only model
Context up to 524,288 tokens
Text out

Good to know: The card reports 77.8 on HMMT Feb 2026 and 72.4 on SWE-bench Verified for this model. Other people have published 48 quantized and 16 fine-tuned versions.

Title: HMMT Feb 2026 score
K2-Horizon-7B: 77.8 points
Granite 4.2-8B: 66.5 points
Qwen3.5-9B: 65.7 points
Gemma 4-12B: 63.1 points

• License: Apache 2.0.
• Cost to run: none named.
• Data it reaches: the files come from Hugging Face, and the model then runs on your own hardware.
• Not for: languages beyond English, since Hugging Face lists only English.

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 9B
• Weights: about 18 GB in BF16; about 4.5 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 42,327
• Likes: 287
• License: apache-2.0
• Languages: en
• Quantized versions: 48
• Fine-tunes: 15
• Tags: transformers, safetensors, k2_horizon, text-generation, k2-horizon, 7b, dense, open-weights, conversational

Links

• Model: https://huggingface.co/IFM/K2-Horizon-7B
• Files: https://huggingface.co/IFM/K2-Horizon-7B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download IFM/K2-Horizon-7B

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