IFM releases K2 Horizon fleet of six open-weight models (0.9B–375B); Meta FAIR introduces research preference models to rank GPU experiments

The Institute of Foundation Models released K2 Horizon, a family of six open-weight models spanning 0.9B, 3.7B, 7B, 32B, 36B, and 375B parameters under the Apache 2.0 license (MarkTechPost). The wider release strategy contrasts with typical single-checkpoint model launches and includes accompanying benchmarks across the scale range.

Meta FAIR, working with Oxford and UCL researchers, introduced AI Research Preference Models (RPMs), a method for ranking proposed ML experiments before execution to optimize GPU resource allocation (MarkTechPost). The frozen LLM-based judges rank up to 15 unexecuted candidates and select one to run, improving average normalized scores from 0.684 to 0.729 on AIRS-Bench. Additionally, H Company released NeoMME, a family of 260M and 800M single-tower multimodal encoders that process multilingual text and raw image patches without separate vision towers (MarkTechPost).