General-Purpose LLMs

The useful comparison is not which model is smarter. Frontier labs build general intelligence, and they are very good at it. The question for legal work is what surrounds that intelligence before an answer reaches a client.

ChatGPT and Claude give you a model. HAQQ gives you the legal system built around one: jurisdiction, sources, your firm's own knowledge, permissions, citations and an audit trail. The model does the generation. Almost everything that makes the output usable in a matter happens before and after that step.

A general assistant runs a short loop: your prompt goes to a model, the model returns text. That is the whole product, and its quality is entirely the model's quality on the day.

Justinian® runs a longer one. It reads the question for legal intent, works out which jurisdiction governs and what language that law is written in, pulls the statutes, case law and matter documents that bear on it, picks the model and tools for that kind of work, reasons, checks the output against what it retrieved, attaches the citations, and only then answers.

Each of those steps is a place a generic assistant has no opinion. Asked a question about termination notice, it does not know whether you mean the UAE, France or Ontario, and it will answer confidently either way.

HAQQ routes different parts of a legal task to different models based on what each is good at, rather than sending everything to one foundation model and hoping. A frontier model still performs the final generation, and we say so plainly.

What that buys is independence. Orchestration, retrieval, jurisdiction logic, memory and safety are ours and run the same whichever model sits underneath. When a better or cheaper model ships, adopting it is a configuration change rather than a rebuild, and the legal layer that took years to build carries over intact.

It also means frontier progress is a tailwind rather than a threat. Better models make HAQQ better without making HAQQ redundant, because the part that understands law is not the part being replaced.

Legal work carries the most sensitive material a person or a business has: disputes, client files, internal documents, strategy. That is a different risk profile from asking a chatbot to summarise an article, and it is not solved by a setting.

So encryption, access control, cross-matter isolation, document anonymisation before processing and data residency are part of how the system is built rather than features around it, and the HAQQ terms governing a deployment exclude customer data from model training. It matters here because people do not simply ask a legal assistant general questions. They upload the entire matter.

HAQQ is built for lawyers and for people who are not lawyers, and those two readers need different answers to the same legal question.

A lawyer wants the authorities, the reasoning, the risk analysis and draft language they can work from. Someone reading their own employment contract wants plain language, what it means for them, and what to worry about. The underlying analysis is the same; what changes is how it is expressed.

A general assistant flattens that distinction, because it has no idea which of the two it is talking to.