THE AUTHOR:
Ayushman Dash, Head of Data and AI at Jus Mundi
A good associate doesn’t hand over the first answer they find. They check what they have, spot what’s missing, go back to the sources, and only then write it up.
Jus AI 3 now works that way and advances that standard further than we ever have before. It is a new generation of the research engine that powers every mode, and it changes something fundamental about how Jus AI works: before it answers, Jus AI 3 now assesses what it has found, identifies what is missing, and goes back for more by running targeted follow-up research or verification. When a question is straightforward, it skips unnecessary reasoning and gets to the answer faster. When the question is complex, it goes deeper before it responds.
The result is research that is more complete, grounded in more authoritative sources, and delivered in the structure and format best suited to the user’s request.
Jus AI 3: A Different Kind of Research Engine
Most AI research tools work in a straight line: they gather what they can, reason through it, and produce an answer. That works well for focused, well-defined questions. But arbitration research rarely works that way. A question about a treaty standard might require weighing awards that interpret it differently, commentary that explains the divergence, and rules that define the procedural context. Each source points to the next and the research has to follow.
Jus AI 3 is built to do exactly that.
Take a multi-treaty investment claim that pulls across three jurisdictions: each treaty may interpret the same standard differently, and the relevant awards, commentary, and procedural rules sit across a body of material that shifts depending on which thread you follow. Jus AI 3 maps those connections, follows the reasoning from one source to the next, and surfaces what matters, without you having to anticipate every thread at the outset.
This is what adaptive research means in practice.
Underlying all of this is Tenet, our proprietary LLM stack built specifically for international arbitration. Jus AI 3 is the next step in what Tenet makes possible: a research engine that understands how authorities in arbitration connect and how to follow them.
Faster Answers, Stronger Sources, Better Format
Jus AI 3 changes the engine, but what you will feel is the difference in the answers.
Answers that arrive sooner
Answers start streaming sooner than before. For straightforward questions, Jus AI now skips unnecessary reasoning steps and gets to the point directly, meaning less waiting and faster research loops when you are moving quickly through a matter. For complex questions, the additional time goes into the research itself.
Sources you can stand behind
Jus AI 3 more consistently identifies the authorities that matter. It is more precise about which sources are truly relevant to your question, more faithful to what those sources actually say, and makes stronger use of tribunal reasoning. That is often the most important passage in an award, and it is now more reliably surfaced and accurately represented in your answer.
Answers shaped like your work
Jus AI 3 is better at following the structure and format your task calls for. If you upload a draft and ask Jus AI to respond to the opposing party’s arguments, it will answer in the same order. If you are working with a document that uses defined terms and a specific numbering system, it will preserve them. If a table helps, it will use one. The answer fits the task rather than defaulting to a fixed format.
The Results, Measured
We measure every update to Jus AI against the same internal benchmark, evaluating quality across the same dimensions we use across all our modes: relevance, faithfulness, correctness, completeness, and language & format. Jus AI 3 was benchmarked against Jus AI 2 across a sample of 114 prompts, covering the range of queries our users actually ask.
The improvements are consistent across the dimensions that matter most:
- 19% faster time to first answer, meaning answers start appearing sooner and research loops move more quickly.
- 10% increase in perfect relevancy scores, with more answers receiving a 5 out of 5 on relevance, measuring how well the sources retrieved match the prompt and the response.
- 8% increase in perfect faithfulness scores, with more answers receiving a 5 out of 5 on faithfulness, measuring how precisely the sources cited reflect what the response actually states.
Jus AI 2 was already scoring between 4 and 4.6 out of 5 across our quality dimensions, a standard that independent arbitration experts rated as professional-grade. At that level of baseline quality, a 10% increase in perfect relevancy scores or an 8% increase in perfect faithfulness scores is a meaningful step forward on a scale that was already performing at the top end.
Laying the Groundwork for What Comes Next
Jus AI 3 is the most significant upgrade to our research engine to date. It is also the foundation for what we are building next.
Arbitrator selection has always been one of the most consequential decisions in a case, and one of the hardest to research thoroughly. Practitioners have had to move between tools and sources to piece together a picture of how an arbitrator has reasoned, ruled, and operated across matters. No tool has connected that research into a single, smooth workflow.
That changes with what we are building next. For the first time, Jus AI will be the starting point for arbitrator search (identify practitioners who match case and appointment criteria) and arbitrator due diligence (take a deep dive into named profiles before nomination, acceptance, or objection). Coming this quarter, it combines conversational AI research with the depth of Jus Mundi and Jus Connect data in one experience. You will be able to research an arbitrator, explore their reasoning across awards, assess patterns in their decisions, and move from discovery to due diligence without switching tools. For example, users will be able to ask ‘Provide potential presiding arbitrators fluent in Spanish for an ICC commercial mining dispute involving a Latin American party, seated in Paris.’
Jus AI 3’s adaptive research is what makes that possible: an engine that can follow connections across a large and complex body of material, surface what is relevant, and present it in the structure the task demands.
The goal has always been the same: arbitration intelligence that matches the complexity of the work. Jus AI 3 moves us further towards that standard.
Arbitration Research, at Its Sharpest
Arbitration research has always demanded more: more precision, more depth, more confidence in the answer before the work begins. Jus AI 3 raises the standard again, with a research engine that adapts to the question, finds the authorities that matter, and delivers the answer in the form your task actually needs.
Jus AI 3 is live now and available to every Jus AI user. Try it now.
About Jus Mundi
Founded in 2019 and recognized as a mission-led company, Jus Mundi is a pioneer in the legal technology industry dedicated to powering global justice through artificial intelligence. Headquartered in Paris, with additional offices in New York, London, and Singapore. Jus Mundi serves over 150,000 users from law firms, multinational corporations, governmental bodies, and academic institutions in more than 90 countries. Through its proprietary AI technology, Jus Mundi provides global legal intelligence, data-driven arbitration professional selection, and business development services.
*The views and opinions expressed by authors are theirs and do not necessarily reflect those of their organizations, employers, or Daily Jus, Jus Mundi, or Jus Connect.
