post-training
Everything done to a model after the expensive base training, to shape how it behaves.
Building a language model happens in two broad phases. First comes pre-training, where the model reads an enormous amount of text and learns how language works. That is the part that costs hundreds of millions and only a handful of organisations can afford. Post-training is everything after that: teaching the model to follow instructions, to be helpful and safe, or to specialise in medicine, law or code.
Post-training is comparatively cheap, which is why open-weight models matter so much. If someone publishes a strong base model, a smaller company can post-train it on its own domain data and end up with a genuine specialist. Harvey did exactly that to build Harvey Tenet on top of Kimi K3.
-
A Legal AI Company Built Its Own Model, and It Started From an Open Chinese One
-
GLM-5.3 Got Much Better at Breaking Software, So Z.ai Is Holding the Weights Back
-
AMD trained a fully open model on its own chips and published everything except a commercial licence
-
OpenAI Says GPT-5.6 Sol Trained Its Own Smaller Sibling