Release2026-09-02

IFM published its K2 Horizon family of open text-generation models on Hugging Face. The line-up spans small models of about 1B parameters up to a 375B mixture-of-experts model. A 36B mixture-of-experts variant and 32B, 7B and 3.7B dense models sit in between. IFM also released FP8 and GGUF versions, which are compressed builds that run on less memory. Two variants carry an "Uno" label. Alongside the models, IFM posted five datasets covering pretraining text, code reasoning, maths reasoning and instruction tuning. The largest, TxT360-v2, holds about 1.84 billion rows. The collection page gives no benchmark scores, licence terms or training details.

What it unlocks

Downloading an open family of K2 Horizon models in sizes from about 1B to 375B, with quantized builds.

  • sizes from 0.9B to 375B parameters
  • 375B model with 23B active
  • 5 datasets published

What you need to act on it

  • hardware or a hosting provider to run the weights

Send this to someone who needs it

Shares the story and its sources. Nothing about you.

What does this mean for your job?

This is the story as everyone gets it. Once a week we send you the version written for your role — what changed, why it matters for the work you actually do, and one thing to try. Free while we tune it.