IFM/K2-Horizon-7B

Hugging Face
TEXT GENERATIONPricing:Input $0.3 / Output $1.2Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kPublished:Sep 1, 2026License:apache-2.0Architecture:Transformer0.2K Open Weights Warm

IFM/K2-Horizon-7B is a 9 billion parameter decoder-only language model developed by IFM, featuring a native 512K context window. This model excels in agentic, coding, long-context, and reasoning benchmarks, demonstrating strong performance in areas like competition mathematics (73.3% on HMMT Feb 2026) and software engineering (70.6% on SWE-bench Verified). It is optimized for complex problem-solving and agentic tool use, making it suitable for advanced AI applications requiring deep reasoning and extensive context understanding.

Loading preview...

K2-Horizon-7B: A 9B Parameter Model with 512K Context

K2-Horizon-7B, developed by IFM, is a 9 billion parameter decoder-only model distinguished by its exceptionally large native 512K token context window. This model is designed for robust performance across a variety of demanding tasks, including agentic operations, coding, and complex reasoning.

Key Capabilities and Differentiators

  • Massive Context Window: Features a native 524,288-token context, enabling deep understanding and generation for long-form content and complex interactions.
  • Strong Performance Baselines: Achieves competitive results on benchmarks, including 73.3% on HMMT Feb 2026 for competition mathematics, 70.6% on SWE-bench Verified for software engineering, and 68.0% on LCR for long-context reasoning.
  • Agentic and Coding Prowess: Demonstrates strong capabilities in agentic tool use (25.8% on tau3-Banking) and coding tasks (31.6% on SciCode, 39.1% on Terminal-Bench 2.1).
  • Open and Transparent: IFM provides fully open training data, recipes, code, and evaluation resources, fostering transparency and research.
  • Optimized Inference: Incorporates Diffusion Adapters for faster inference and supports specific reasoning and tool-call parsers for enhanced functionality.

When to Use This Model

K2-Horizon-7B is particularly well-suited for use cases requiring:

  • Advanced Reasoning: Applications demanding high-level mathematical, scientific, or general reasoning capabilities.
  • Long-Context Understanding: Tasks that benefit from processing and generating very long texts, such as document analysis, summarization, or extended dialogue.
  • Code Generation and Software Engineering: Development workflows involving code generation, debugging, or automated software engineering tasks.
  • Agentic Applications: Building AI agents that can effectively use tools, browse the web, or interact with complex environments.

For optimal performance, IFM recommends using specific settings like reasoning_effort="high", temperature=1.0, top_p=0.95, and allowing at least 32,768 output tokens to prevent truncation of reasoning.