McCarthy Building Companies presented the underlying scores at DevCon 7: after tuning a 4 billion parameter model and running it on its evaluation set, it reached about 73% balanced accuracy compared with 62% for Astra. The company said the result came at a fraction of the cost of the frontier model and with 20% fewer false alerts. The model factory behind the result is built on Ontology-linked data, Model Studio, GPU Hubs and AIP Evals, turning construction domain expertise into a fine-tuned model. The comparison rests on McCarthy's own evaluation set, and no cost figures were given.

AI summary of the linked source β€” not editorial content.

Loading related coverage…

Comments

Loading comments…