Open source AI model release

Source: https://www.instagram.com/reel/Dc5KWc8SgAg/?stkn=MXh6M3BsdjdoZzkzZQ==

AI generated summary of the linked source. Posted anonymously. Transcripts are published only for public links, never for uploads, documents or pasted text.

Your library

The AI link is plain markdown. Paste it into any assistant and it can read the whole public entry.

Summary

A group called the Institute of Foundation Models just released K2 Horizon, an open source family of six AI models ranging from 0.9 billion to 375 billion parameters, all under Apache 2.0. Unlike most open releases that only hand you finished model weights, this one exposes the entire training pipeline, including data, code, and checkpoints, letting anyone reproduce or customize how the model learned.

Core ideas

One context card plus 6 ideas

Full stack transparency: K2 Horizon releases training data, training code, configs, intermediate checkpoints, and raw logs, not just final weights, so developers can trace exactly how reasoning and tool use emerged during training.
Six sizes for every use case: Models range from a 375 billion parameter version for enterprise workloads down to a 0.9 billion parameter model small enough for a smartwatch or smart glasses.
No licensing friction: Everything ships under Apache 2.0, meaning developers can build, ship, and sell products using it without paying license fees or asking permission.
New efficiency trick called Mova: This attention architecture lets a 36 billion parameter model run using only about 4 billion active parameters per token, cutting compute cost without shrinking the model's total knowledge.
Sparse activation explained simply: Instead of firing all 36 billion parameters for every word, Mova routes each word to a small subset of specialist parameters, roughly 4 billion, making inference cheaper and faster.
Reproducibility as the real value: The ability to watch when specific capabilities like reasoning appeared during agentic post-training turns this release into a research tool, not just a product.

Quotes

“It's called K2 Horizon built by the Institute of Foundation Models out of Abu Dhabi and it's not one model it's a whole fleet.”
@v.i.s.h.ai

“Almost every open model you've used only gives you the final weights, you get the finished cake never the recipe.”
@v.i.s.h.ai

“So Mova breaks that chain, instead of waking up every dial the model keeps all 36 billion on standby and only routes each word to the small group of specialists that actually matter for it.”
@v.i.s.h.ai

“The smarter you make a model the more you pay every time it opens its mouth.”
@v.i.s.h.ai

Resources

Links to the tools, products, repos or reading this entry mentions. Found by a web search of what the entry names, so you can go and use them.

K2 Horizon
rits.shanghai.nyu.edu
Official IFM page introducing and describing the K2 Horizon family of six open-source AI models.
Institute of Foundation Models
mbzuai.ac.ae
Canonical product page for IFM’s K2 Horizon model family
Mova
huggingface.co
Official IFM blog page introducing K2 Horizon and its MoVA sparse attention architecture

Remix this knowledge

Turn this entry into something you can post. Pick a format and get a fresh, original rewrite.

Remixes are AI generated, original wording, and free to reuse.

Talk it through

Open a round table on this entry and argue with other people about what it means.

Start a round table

Drop your own

Paste any video, podcast or audio link and get the transcript, the core ideas and the quotes back in one pass.

Try Drop