K2 Horizon

K2 Horizon

Institute of Foundation Models
+
+

Related Products

  • Gemini Enterprise Agent Platform
    999 Ratings
    Visit Website
  • Google AI Studio
    30 Ratings
    Visit Website
  • LM-Kit.NET
    29 Ratings
    Visit Website
  • LTX
    182 Ratings
    Visit Website
  • Concord
    237 Ratings
    Visit Website
  • RentGuruz
    8 Ratings
    Visit Website
  • RouteGenie
    49 Ratings
    Visit Website
  • Descartes Fleet Performance Management
    28 Ratings
    Visit Website
  • Planview AdaptiveWork
    714 Ratings
    Visit Website
  • ManageEngine Endpoint Central
    3,286 Ratings
    Visit Website

About

K2 Horizon is a connected fleet of six open models spanning 375B-A23B, 36B-A4B, 32B, 7B, 3.7B, and 0.9B, designed to deliver strong performance across reasoning, mathematics, coding, agentic tasks, and general capabilities. The models share core architecture, vocabulary, training methodology, interfaces, evaluation infrastructure, and deployment tooling, making it easier to move between sizes and route workloads dynamically. The 375B-A23B model is the fleet’s most capable option for complex reasoning, software engineering, research, and long-horizon agentic work, while the 32B and 36B-A4B models target powerful local deployment. The 36B-A4B model introduces Mixture-of-Value Attention, combining sparse attention with Mixture-of-Experts layers to activate about 4 billion parameters per token while approaching the performance of the dense 32B model.

About

Muse Glimmer is a 30-billion-parameter open-weights model from Meta Superintelligence Labs, optimized for always-on local agent workflows. Small enough to run on a Mac or PC with a single consumer GPU, it is designed for local agents, function calling, coding, and LLM-as-a-judge evaluation without depending on cloud infrastructure or network access. The model combines long-horizon execution, precise tool calling, multimodal understanding, long-context memory, and instruction following. It can complete end-to-end agentic tasks, sustain multi-step reasoning across extended workflows, recover from failed or unexpected tool calls, and accept interleaved text and images through a dedicated perception encoder for interpreting screenshots, charts, and documents. Muse Glimmer works with OpenClaw and other agentic orchestration patterns, supports controllable reasoning effort, and is trained on data from more than 100 languages.

Platforms Supported

Windows
Mac
Linux
Cloud
On-Premises
iPhone
iPad
Android
Chromebook

Platforms Supported

Windows
Mac
Linux
Cloud
On-Premises
iPhone
iPad
Android
Chromebook

Audience

AI researchers, developers, and organizations searching for open models for reasoning, coding, tool use, agentic workflows, and deployments

Audience

Developers building private, always-on AI agents that need strong reasoning, tool use, and multimodal capabilities on local hardware

Support

Phone Support
24/7 Live Support
Online

Support

Phone Support
24/7 Live Support
Online

API

Offers API

API

Offers API

Screenshots and Videos

Screenshots and Videos

Pricing

No information available.
Free Version
Free Trial

Pricing

Free
Free Version
Free Trial

Reviews/Ratings

Overall 0.0 / 5
ease 0.0 / 5
features 0.0 / 5
design 0.0 / 5
support 0.0 / 5

This software hasn't been reviewed yet. Be the first to provide a review:

Review this Software

Reviews/Ratings

Overall 5.0 / 5

Pros & Cons from Real Users

Pros

  • Muse Glimmer is exciting because it brings serious AI capability closer to the device. An open-weight model that can run on a laptop or desktop is a big deal for developers, builders, and AI power users who want more control. I like that it is focused on agentic tasks, not just basic chat. If it can handle reasoning, coding help, workflow automation, and local experimentation well, it could be really useful for private projects and always-on agents. The open-weight angle is the biggest win. Being able to download, modify, and run the model locally makes Muse Glimmer feel much more flexible than a closed API-only model.

Cons

  • I would still want to test it hard before trusting it for serious work. Smaller local models can be impressive, but they still need to prove themselves on coding, tool use, long tasks, and messy real-world prompts. Running locally also means the experience depends on your hardware. Even if it works on consumer devices, performance, speed, and setup may vary a lot.

Training

Documentation
Webinars
Live Online
In Person

Training

Documentation
Webinars
Live Online
In Person

Company Information

Institute of Foundation Models
Founded: 2025
United States
ifm.ai/blog/k2/

Company Information

Meta
Founded: 2004
United States
meta.ai/

Alternatives

LongCat-2.0

LongCat-2.0

LongCat

Alternatives

Grok 4.6

Grok 4.6

SpaceXAI
MiMo-V2.5-Pro

MiMo-V2.5-Pro

Xiaomi Technology
Ling 3.0 Flash

Ling 3.0 Flash

Ant Group
Command A+

Command A+

Cohere AI
Hy4

Hy4

Tencent

Categories

Categories

Integrations

ExecuTorch
Hermes Agent
Hugging Face
LM Studio
Meta AI
Muse Code
Ollama
OpenClaw
OpenCode
Unsloth

Integrations

ExecuTorch
Hermes Agent
Hugging Face
LM Studio
Meta AI
Muse Code
Ollama
OpenClaw
OpenCode
Unsloth
Claim K2 Horizon and update features and information
Claim K2 Horizon and update features and information
Claim Muse Glimmer and update features and information
Claim Muse Glimmer and update features and information