LocalClaw/LLM/K2-Horizon-MoVA-36B-A4B
Open-weight MoE

K2-Horizon-MoVA-36B-A4B

Catalogue summary: IFM Apache 2.0 sparse K2 Horizon release with Mixture-of-Experts plus Mixture-of-Values attention, 36B total / 4B active parameters, 512K context and an official BF16 GGUF path for high-memory local workstations.

Repository editorial metadata; verify comparative claims in the linked upstream material.

96 GB catalogue minimum96 GB RAMBF16 GGUFCoding assistant
Choose an app

This model needs a special runtime. Unsupported apps are clearly marked.

Compare models
LM StudioNot available for this model
UnslothNot available for this model
OllamaNot available for this model
Open on Hugging FaceFiles, licence and available downloads
llama.cppNot available for this model
Use with LocalClawOptional workspace after the model is installed
This model needs its official runtimeK2 Horizon llama.cpp fork / pending upstream llama.cpp support
Parameters
36B (4B active, MoE)
Minimum RAM
96 GB
Model size
72 GB
Quantization
BF16 GGUF

Can K2-Horizon-MoVA-36B-A4B run locally?

K2-Horizon-MoVA-36B-A4B has a catalogue minimum of 96 GB RAM with BF16 GGUF. Actual memory use and speed vary by context length, runtime, backend and system headroom.

Use the official K2 Horizon llama.cpp fork / pending upstream llama.cpp support setup. The current low-bit files are not a stock LM Studio install.

Model sourcePublic GGUF repository (verified 2026-09-07): IFM/K2-Horizon-MoVA-36B-A4B-GGUFUpstream releasePaper or model cardLicence text
chatcodereasoningagenticlong-contextpowergeneral

Install path

01
Check RAM fitMinimum 96 GB RAM. Start with the BF16 GGUF quant.
02
Load the modelFollow the official K2 Horizon llama.cpp fork / pending upstream llama.cpp support instructions. Stock LM Studio support is not confirmed.
03
Control locallyUse LocalClaw to manage models, agents, chat, channels and scheduled OpenClaw work.

Catalogue record

  • Family: k2-horizon
  • Parameters: 36B (4B active, MoE)
  • Recommended quantization: BF16 GGUF
  • Catalogue minimum RAM: 96 GB
  • Catalogue model size: 72 GB
  • Tags: chat, code, reasoning, agentic, long-context, power, general

Practical limits

  • Catalogue RAM is a minimum estimate, not a guarantee for every context length or runtime.
  • Speed and memory use vary by quantization, backend, context length and system headroom.
  • Verify architecture, licence and usage restrictions in the linked upstream material before deployment.

Catalogue tags

  • chat
  • code
  • reasoning
  • agentic
  • long-context
  • power

Capability profile

Repository catalogue ratings used by LocalClaw's editorial rubric. They are not a standardized third-party benchmark.

speed
5
quality
9
coding
9
reasoning
9

Technical notes

Developer
IFM
License
Apache 2.0
Context window
524,288 tokens
Architecture
Sparse K2 Horizon Mixture-of-Experts model with Mixture-of-Values attention, 36B stored parameters, about 4B active parameters per token and a 512K-token context window.

This model fits these next steps

Hardware fit is based on LocalClaw's RAM tier, model size and quantization metadata. Always leave memory headroom for your OS and runtime.

Large unified memoryMac Studio M4 Max 128GB128GB AI PC classNVIDIA GB10 / DGX SparkHigh-memory picks128GB RAM guide

Related catalogue entries

Linked mechanically by family, shared tags and nearby RAM tier; this is not a quality ranking.

K2-Horizon-0.9B 0.9BNemotron 3.5 Lightning 30B-A3B 30B (3B active, MoE)Ornith-1.5-35B-A3B 35B (3B active, MoE)Mistral Small 4 (119B-A6.5B) 119B (6.5B active, MoE)Qwen 3.6 35B-A3B 35B (3B active, MoE)Qwen3.8-27B 27BDeepSeek V4 Flash Vision Exp 284B (13B active, multimodal MoE)Ling Flash Base 2.0 106B (6.1B active, MoE)

Where to go next