---

pipeline_tag: text-generation

base_model:

• IFM/K2-Horizon-7B

library_name: llama-cpp

tags:

• k2-horizon
• 7b
• dense
• open-weights
• ifm
• gguf
• nvfp4

language:

• en

license: apache-2.0

license_name: apache-license-2.0

license_link: https://apache.org/licenses/LICENSE-2.0

---

K2-Horizon-7B NVFP4 — GGUF

NVFP4 GGUF conversions derived from the K2-Horizon-7B model family. This repository provides NVFP4 (NVIDIA 4-bit floating-point) GGUF files, making the model usable with llama.cpp and other GGUF-compatible inference engines.

Model Overview

K2-Horizon-7B is the medium dense member of the K2-Horizon family: a 7B-core decoder-only model with a 512K context window. It is a strong dense baseline evaluated across agentic, coding, long-context, and reasoning benchmarks.

Model Architecture

| Property | Value |

| :----------------------- | :-------------------- |

| Architecture | K2-Horizon-7B |

| Parameters | 7B |

| Context Length | 512K tokens (524,288) |

| Supported Modalities | Text |

| Type | Dense |

Benchmark Results

| Benchmark | K2-Horizon-7B | Reference 1 | Reference 2 | Reference 3 |

| :----------------------- | :-----------: | :----------------------: | :--------------------: | :---------------------: |

| Math | | | | |

| HMMT Feb 2026 | 73.3 | 63.1 (Gemma 4-12B) | 65.7 (Qwen3.5-9B) | 66.5 (Granite 4.2-8B) |

| Coding | | | | |

| SWE-bench Verified | 70.6 | 30.6 (Gemma 4-12B) | 47.7 (Granite 4.2-8B) | 50.8 (Qwen3.5-9B) |

| Scientific Reasoning | | | | |

| HLE | 18.6 | 9.7 (Granite 4.2-8B) | 14.9 (Qwen3.5-9B) | 15.7 (Gemma 4-12B) |

| SciCode | 31.6 | 27.5 (Qwen3.5-9B) | 28.0 (Mistral Small 4) | 30.4 (Granite 4.2-8B) |

| General | | | | |

| LCR | 68.0 | 43.3 (Granite 4.2-8B) | 61.7 (Gemma 4-12B) | 65.3 (Qwen3.5-9B) |

| Coding | | | | |

| Terminal-Bench 2.1 | 39.1 | 18.4 (Granite 4.2-8B) | 27.3 (Gemma 4-12B) | 29.2 (Qwen3.5-9B) |

| Agents | | | | |

| tau3-Banking | 25.8 | 7.0 (Qwen3.5-9B) | 7.6 (Granite 4.2-8B) | 24.0 (Muse Glimmer-30B) |

| BrowseComp | 59.0 | 53.5 (DeepSeek V4 Flash) | 54.9 (GPT-5) | 56.6 (LongCat Flash) |

Scores in %. Bold marks the best score in each row. Benchmark results are sourced from the original K2-Horizon-7B model description.*

GGUF Files

| File | Format | Description |

| :------------------------- | :----- | :-------------------- |

| K2-Horizon-7B-nvfp4.gguf | NVFP4 | NVFP4 quantized model |

Usage

llama.cpp (CLI)

# Run inference
./llama-cli \
  -m K2-Horizon-7B-nvfp4.gguf \
  --temp 1.0 --top-p 0.95

llama-server (OpenAI-compatible API)

# Start the server
./llama-server \
  -m K2-Horizon-7B-nvfp4.gguf \
  --port 8080

Key Features

• Strong dense baseline. A 7B-class dense model evaluated across agentic, coding, long-context, and reasoning benchmarks.
• 512K context. Native 524,288-token context from the midtraining stages onward.
• NVFP4 Quantization. Optimized for NVIDIA Blackwell and compatible hardware.

Citation

@misc{k2horizon2026,
  title  = {Introducing K2 Horizon: Frontier Performance, Radically Open},
  author = {{IFM Team}},
  year   = {2026},
  url    = {https://ifm.ai/blog/k2/},
}