IFM Releases K2 Horizon: Six Apache 2.0 Models From 0.9B to 375B
The Institute of Foundation Models (IFM), the frontier AI research lab launched by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in May 2025, has announced K2 Horizon—a significant release that departs from industry convention by offering six distinct model checkpoints rather than a single offering. This multi-tiered approach demonstrates a strategic shift in how foundation model developers approach open-source AI deployment, providing options across computational scales from 0.9 billion to 375 billion parameters.
The K2 Horizon family comprises six models with varying parameter counts: 375B, 23B, 36B, 32B, 7B, 3.7B, and 0.9B variants. Each model is released under the Apache 2.0 license, emphasizing accessibility and commercial usability. This scaled-fleet approach contrasts sharply with typical industry practice, where organizations typically release a single flagship model alongside benchmark comparisons. By providing multiple checkpoints, IFM enables developers and enterprises to select models appropriate for their specific computational constraints and performance requirements.
The release strategy reflects growing recognition that foundation model deployment varies significantly across use cases, from edge computing applications requiring minimal parameters to data center environments capable of supporting massive models.
- Democratized AI Access: Multiple model sizes make advanced AI capabilities available to organizations with varying hardware infrastructure and budgets
- Computational Flexibility: Developers can optimize for latency, throughput, and resource consumption without sacrificing capability
- Open-Source Competition: Apache 2.0 licensing encourages commercial adoption and integration into proprietary systems
- Frontier Lab Positioning: IFM's inaugural release establishes MBZUAI as a credible competitor in open-source foundation model development
- Industry Standardization: The multi-checkpoint approach may establish new expectations for foundation model releases
This release signals a fundamental evolution in foundation model distribution strategy. Rather than forcing developers to choose between limited options, IFM's comprehensive family addresses real-world deployment diversity. For enterprises, researchers, and developers, having six validated models spanning two orders of magnitude in parameter count provides genuine flexibility. The Apache 2.0 licensing further reduces barriers to adoption, positioning K2 Horizon as a significant contribution to the open-source AI ecosystem during a critical period of foundation model proliferation.
Key Takeaways
- The Institute of Foundation Models (IFM), the frontier AI research lab launched by Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in May 2025, has announced K2 Horizon—a significant release that departs from industry convention by offering six distinct model checkpoints rather than a single offering.
- This multi-tiered approach demonstrates a strategic shift in how foundation model developers approach open-source AI deployment, providing options across computational scales from 0.
- 9 billion to 375 billion parameters.
- The K2 Horizon family comprises six models with varying parameter counts: 375B, 23B, 36B, 32B, 7B, 3.
Read the full article on MarkTechPost
Read on MarkTechPost