IFM's K2 Horizon aims to make the whole training pipeline inspectable

kimmonismus · x · 2026-09-09

IFMAI launched K2 Horizon, a connected fleet of six models ranging from 0.9B to 375B total parameters. The ambitious part: making the training process inspectable across pretraining, reasoning, and agentic post-training, with sparse variants activating only 4B and 23B parameters per token.

Related event: IFM Launches K2 Horizon Fleet Spanning 0.9B to 375B Parameters(2 posts)→

Original post →

More from Models

Models channel →