# K2-Horizon-MoVA-36B-A4B ![Organization](https://img.shields.io/badge/Org-IFM-555) ![Context](https://img.shields.io/badge/Context-512K%20tokens-1f6feb) ![Pricing](https://img.shields.io/badge/Price-Open%20weights-2da44e) ![License](https://img.shields.io/badge/License-Apache%202.0-d29922) ![Modality](https://img.shields.io/badge/Modality-Text-8957e5) ![Released](https://img.shields.io/badge/Released-Sep%202026-555) > **Quick answer:** K2-Horizon-MoVA-36B-A4B is IFM's sparse Mixture-of-Experts model using Mixture-of-Values attention (MoVA) — 36B total parameters, only 4B active per token, with a native 512K-token context window. It outscores open-weight dense and MoE models up to 15x its active-parameter size on agentic and reasoning benchmarks, and is competitive with closed frontier models. Open weights, Apache 2.0. ## At a Glance **Where K2-Horizon-MoVA-36B-A4B leads** - Best-in-class efficiency: frontier-adjacent agentic scores at only 4B active parameters - Strong Terminal-Bench 2.1 and τ³-Banking scores versus much larger open peers - 512K native context; fully open weights **Where it lags** - GPQA Diamond and AA-Omniscience trail larger dense/MoE peers like Gemma 4 31B and Nemotron 3 Ultra - CritPt (frontier physics) remains a weak point across the whole K2-Horizon family **Best for:** cost-sensitive deployments needing strong agentic tool-use and terminal capability without the inference cost of a 30B+ active-parameter model. ## What K2-Horizon-MoVA-36B-A4B Is K2-Horizon-MoVA-36B-A4B is the efficiency-focused member of IFM's K2-Horizon family. It uses a novel Mixture-of-Values attention (MoVA) mechanism paired with sparse MoE routing to store 36B parameters while activating just 4B per token — a ~9x sparsity ratio. IFM's own framing is that at this activated-parameter budget it competes with models up to 15x larger, positioning it as an efficiency play for agentic and reasoning workloads rather than a raw-parameter-count flagship. ## Specifications | Field | Value | |---|---| | Organization | IFM | | Parameters | 36B total / 4B active | | Context window | 524,288 tokens (512K) | | Architecture | Sparse MoE with Mixture-of-Values attention (MoVA) | | License | Apache 2.0 | | Release date | 2026-09 | | Modality | Text | ## Pricing | | Input (per 1M tokens) | Output (per 1M tokens) | |---|---|---| | Open weights | — | — | Open weights: free to download and self-host. Validated SGLang recipe: 2x H200, TP=2, EP=2. ## Context Window K2-Horizon-MoVA-36B-A4B has a **524,288-token (512K) context window**, native from the midtraining stages onward — enabling long-document and long-horizon-agent workloads at a fraction of the active-parameter cost of its dense-context peers. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [τ³ Banking](/benchmarks/tau-bench/tau3-banking) | 26.8 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B) | 2026-09 | | [TerminalBench 2.1](/benchmarks/tbench/terminalbench-21) | 58.6 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B) | 2026-09 | | [SciCode](/benchmarks/scicode-team/scicode) | 38.9 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B) | 2026-09 | | [Humanity's Last Exam](/benchmarks/scale-ai/humanitys-last-exam) (no tools) | 25.2 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B) | 2026-09 | | [GPQA Diamond](/benchmarks/rein-et-al/gpqa-diamond) | 80.8 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B) | 2026-09 | | [CritPt](/benchmarks/critpt-team/critpt) | 2.1 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B) | 2026-09 | | [AA-LCR](/benchmarks/artificial-analysis/aa-lcr) | 66.3 | [IFM model card](https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B) | 2026-09 | *AA-Omniscience Accuracy (18.8) and Non-Hallucination (69.2) were also reported but use a different metric structure than the platform's combined AA-Omniscience Index — pending mapping decision, not yet added. Scores above are reported by IFM and shown for context; not Benchgen measurements.* ## Use K2-Horizon-MoVA-36B-A4B via API ```python from openai import OpenAI client = OpenAI(base_url="http://localhost:30000/v1", api_key="EMPTY") response = client.chat.completions.create( model="IFM/K2-Horizon-MoVA-36B-A4B", messages=[{"role": "user", "content": "Explain the result step by step."}], temperature=1.0, top_p=0.95, extra_body={"chat_template_kwargs": {"reasoning_effort": "high"}}, ) print(response.choices[0].message.content) ``` ## Frequently Asked Questions

What is K2-Horizon-MoVA-36B-A4B?

It's IFM's efficiency-focused Mixture-of-Experts model using Mixture-of-Values attention (MoVA): 36B total parameters, only 4B activated per token, with a 512K-token context window.

What is K2-Horizon-MoVA-36B-A4B's context window?

524,288 tokens (512K).

How much does K2-Horizon-MoVA-36B-A4B cost?

Open weights — free to download; cost is hosting/inference only.

Is K2-Horizon-MoVA-36B-A4B open source?

Yes, released under the Apache 2.0 license.

--- *Specs and scores sourced from IFM's official Hugging Face model card; third-party benchmark scores attributed inline. Last updated 2026-09-03.*

K2 Horizon MoVA 36B A4B — Training jobs

Training jobs