OPT-175B

OPT-175B, Open Pre-trained Transformer, was Meta AI's May 2022 bid to open up frontier-scale language model research. At announcement it was the first 175-billion-parameter language model made available to the broader research community, trained with a publicly documented logbook describing the training process day by day. Meta released the model family from 125M to 175B parameters to academic researchers, government labs, civil society groups and industry research labs, arguing that restricted access was hindering work on robustness, bias and toxicity.

The release came with a deliberate caveat that the catalog flags honestly: 'open' here did not mean OSI-open. OPT-175B shipped under a non-commercial license agreement restricted to research use, and Meta explicitly said it was not for production deployment. Even so, it gave researchers outside the handful of frontier labs their first look inside a GPT-3-scale model. The weights remain available for research under those terms, so openWeights is true with the license restriction noted.