IFM open-sources a six-model AI fleet

The Institute of Foundation Models released K2 Horizon on September 3, a fleet of six models from 0.9B to 375B parameters with weights, code, training data, and methodology all published under Apache 2.0. IFM calls it the largest fully open model release to date, and unlike most “open weight” launches, such as Moonshot’s Kimi K3, it includes the training data and recipes needed to actually reproduce the models, not just run them.

The models share a core architecture, vocabulary, and deployment tooling across sizes (the 0.9B variant uses a smaller vocabulary), and IFM claims the 0.9B, 3.7B, and 7B models set new state of the art for their size class on reasoning, math, coding, and agentic-task benchmarks. The smallest model targets constrained hardware like watches and glasses; the 3.7B and 7B sizes target phones and other on-device deployments.

For operators, full data and methodology transparency matters more than the parameter count: it’s the difference between a model you can audit and fine-tune with confidence and one you’re trusting on the vendor’s word. If you run anything at the small end of the stack, on-device or edge, this is a fleet worth benchmarking against your current default before your next model-selection review, alongside other agent-focused open releases like Qwen’s open agent world model.