General-purpose models have converged. Consultancy firms are now selling similar models to the same companies, and that is their problem. Specialization is where the remaining gains.
Between 2025 and 2026, frontier models from Anthropic and OpenAI crossed the threshold into genuine enterprise reliability, and the large consultancies deployed them at scale.
In recent months, two things have happened. The models stopped meaningfully differentiating from each other. And the deployments stopped clearing their ROI hurdles.
A model trained on everything is trained on nothing in particular. In domains where the vocabulary is precise, the sources are specific, and being current matters â risk, compliance, audit â a general model is working against its own construction.
Earthian built risk language models on that premise: trained on risk, evaluated on risk, and materially ahead of general-purpose models on accuracy and freshness within it. And we have continuously published our benchmarked differences with our partners, including two of the world's largest consultancy firms.
And today we are extending the approach beyond risk. Humanpath is infrastructure for domain experts inside large enterprises to train models on their own data, in their own domain. The expertise required to build a good specialized model, which until now sat inside large frontier labs, is now placed inside these organizations.
We expect the consulting market to reflect this within eighteen months. The firms currently selling the same general-purpose deployments to the same clients will be selling something narrower and more defensible, or they will be competing on price.