US lab unveils ‘fully open’ AI models in challenge to China’s dominance

A US non-profit has released a suite of “fully open” artificial intelligence models – disclosing its training data and pipelines in full – in a bid to challenge Chinese dominance in the open-source AI arena.

The Allen Institute for AI, also known as Ai2, on Thursday released the latest Olmo series of models with their full training data and training pipelines available for public inspection.

The move goes beyond the common practice of Chinese open-source AI developers, who typically make only model weights available for others to view and modify.

Ai2 said the greater transparency could help boost user trust at a time when AI systems were increasingly deployed by institutions to power critical services.

Its new flagship 32-billion-parameter Olmo 3-Think model narrowed the performance gap with leading Chinese models of a similar size, such as Alibaba Cloud’s Qwen3-32B, while training on roughly six times fewer tokens, it said.

Alibaba Group Holding, which owns the Post, operates Alibaba Cloud.