Home / Technology / AI2 Unveils Next-Gen Olmo 3.1 Models
AI2 Unveils Next-Gen Olmo 3.1 Models
13 Dec, 2025
Summary
- New Olmo 3.1 models enhance efficiency, transparency, and control for enterprises.
- Olmo 3.1 Think 32B shows significant gains in math and reasoning benchmarks.
- AI2's Olmo 3.1 Instruct 32B is a capable open-source model for chat and tool use.
The Allen Institute for AI (Ai2) has introduced its advanced Olmo 3.1 family of AI models, emphasizing enterprise-ready efficiency, transparency, and control. These new models build upon previous iterations, with Olmo 3.1 Think 32B showing substantial performance increases in benchmarks for math, reasoning, and instruction-following after extended training.
Complementing the Think variant, Olmo 3.1 Instruct 32B is engineered for sophisticated chat applications, tool integration, and extended dialogues. Ai2 highlights this model as its most competent fully open 32B-scale instruction-tuned model to date, outperforming peers in various evaluations.
Ai2 continues its commitment to open AI development, making Olmo 3.1 models accessible via platforms like Hugging Face, with API access anticipated soon. The institute stresses that the Olmo 3.1 family upholds end-to-end transparency in data, code, and training decisions, advocating for the advancement of both performance and openness in AI.
