Robots Atlas>ROBOTS ATLAS
K2 Horizon
AI Models›K2 Horizon

K2 Horizon

Horizon · Family: K2 Horizon
Fully open family of language models from IFM. MoE variants (375B-A23B, 36B-A4B with MoVA) and dense (32B–0.9B), 512K context, Apache 2.0.
✓ Active✓ Public access⚖ Open sourceLLMReasoning model📁 K2 Horizon
Context window
512K
tokens
Parameters
375B-A23B (flagowy); rodzina 0,9B–375B
parameters
Release date
3 September 2026
🏢Institute of Foundation ModelsProducer
Access:DownloadDeployment:💻 Local☁ Cloud

Overview

K2 Horizon is a family of open language models developed by the Institute of Foundation Models (IFM) — a global research lab affiliated with MBZUAI, growing out of the LLM360 team. The flagship variants were released on September 3, 2026 under the Apache 2.0 license.

Architecture and variants

The family includes Mixture-of-Experts variants and dense models. The flagship K2-Horizon-375B-A23B has 375B parameters, of which 23B are active per token. The K2-Horizon-36B-A4B variant (36B parameters, 4B active) uses IFM’s Mixture-of-Value Attention (MoVA), extending expert routing into the attention layer. Dense models comprise 32B, 7B, 3.7B and 0.9B variants. All feature a native 512K-token context window (524,288).

Openness (LLM360)

Following the LLM360 philosophy, IFM commits to full openness: releasing weights, training code, recipe and training data (partly under licenses such as ODC-BY). Models are also distributed in the GGUF format for local execution.

Results

The flagship 375B-A23B reaches, among others, 87.3% on GPQA Diamond, 32.0% on Humanity’s Last Exam, 70.2% on Terminal-Bench 2.1 and 42.6% on SWE-Bench Pro.

Classification
LLMReasoning model
Family: K2 Horizon
Access & deployment
Download
LocalCloud
Weights: Open source
Key parameters
📏 Context: 512K
🧩 Parameters: 375B-A23B (flagowy); rodzina 0,9B–375B
✓ Tools · ✓ Fine-tuning
📥 Input: text

Technical specification

Context window
512K
tokens
Parameters
375B-A23B (flagowy); rodzina 0,9B–375B
parameters
License
Apache 2.0
Features:✓ Tool use✓ Fine-tuning
Modalities
⬇ Input
text
⬆ Output
textcode

Capabilities and applications

Native model capabilities
Reasoning
The model's ability to reason logically and solve complex problems.
Category: reasoning
Advanced reasoning
The ability to perform multi-step, structured reasoning: analysing problems, planning steps, and drawing conclusions from hypotheses. Reasoning-first models (e.g. GPT-5.1 Thinking) dedicate a portion of inference to chains of thought before responding.
Category: reasoning
Mathematical reasoning
The model's ability to solve mathematical tasks requiring multi-step reasoning — equations, proofs, combinatorics, geometry, calculus and competition-level problems.
Category: reasoning
Coding
Generating, analysing and modifying code in many programming languages. Covers writing functions, debugging, refactoring, code review, and creating tests. Measured by benchmarks such as HumanEval and SWE-bench.
Category: coding
Agentic coding
Multi-hour, multi-step programming tasks performed autonomously by the model: cloning a repository, running tests, iterating on fixes, integrating with CLI tools. Characteristic of Codex variants (GPT-5.1-Codex-Mini, Codex-Max).
Category: coding
Tool use
The model's ability to call external functions, APIs and tools during a conversation: calculator, search engine, code editor, database. The model decides when and how to use a tool and interprets its result.
Category: planning
Long context
Support for large context windows — tens to hundreds of thousands (or millions) of input tokens. Enables analysis of entire codebases, long documents, and many parallel conversations without losing earlier information. GPT-5.1 supports 400,000 tokens.
Category: language
Agentic capability
The model's ability to autonomously plan and execute multi-step tasks by sequentially using tools, maintaining context, and adapting to intermediate results.
Category: planning
Adaptive reasoning effort
The model decides how much 'thinking' to allocate to a given query: simple questions are answered quickly, complex problems receive more inference cycles. A GPT-5.1 feature (both Instant and Thinking) that shortens time on easy tasks and extends it for hard ones.
Category: reasoning

Benchmark results

4 benchmarks
GPQA
accuracy · GPQA Diamond
87.3%%
📄 IFM
K2-Horizon-375B-A23B variant.
Humanity's Last Exam (HLE)
accuracy
32.0%%
📄 IFM
K2-Horizon-375B-A23B variant.
Terminal-Bench 2.0
accuracy · Terminal-Bench 2.1
70.2%%
📄 IFM
K2-Horizon-375B-A23B variant; benchmark v2.1.
SWE-Bench Pro
accuracy
42.6%%
📄 IFM
K2-Horizon-375B-A23B variant.

Technical architecture

Sources and related pages