Arcee AI's Trinity Large release provides checkpoints from different stages of model development. Preview is lightly post-trained and chat-ready, Base is the best pretrainingThe first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.Full definition → checkpoint after the full recipe, and TrueBase is an earlier checkpoint from the same run, taken before instruction data and learning-rate annealing.
That distinction matters for researchers studying what pretraining alone produces. TrueBase exposes a model after 10 trillion text tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → of pretraining without instruction examples, while Base follows the full 17-trillion-token recipe. Preview serves a different purpose: trying the model as an assistant. The checkpoint names describe different experimental starting points, not interchangeable versions.
Terms in this piece · Glossary
pretraining — The first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
