A Northeastern University study uses Olmo’s open weights and accessible activations to identify where demographic bias appears inside a clinical language model, showing how openness can help researchers explain and improve model behavior in high-stakes healthcare settings.
SciArena is used to evaluate foundation models on scientific literature tasks through community-driven, literature-grounded, multidisciplinary reasoning.
Allen Institute for AI (Ai2) releases Olmo-core 3, a framework designed to make large mixture-of-experts language models more efficient to train and scale beyond one trillion parameters. Ai2 says the system processes 52,000 tokens per second on Nvidia B3000 GPUs for a 47-billion-parameter model, about 2.7 times Megatron-core’s reported throughput. It distributes experts, model layers, and optimizer state across GPUs, supports MXFP8, and is available on GitHub for developers and open-source researchers.
