Ivo Legal AI: The Kiwi Contract Startup Going Open Source

Power Points –

This week, at its first user conference in San Francisco, legal AI company IVO did something none of the legal AI heavyweights have been willing to do by releasing a contract-trained AI model that anyone can download, poke at and build on, for free. (The full announcement is here.)

Min-Kyu Jung, a former lawyer at New Zealand big law firm Bell Gully in Auckland, taught himself to code in about two months, quit, (as we wrote in LawFuel 18 months ago) and started his legal AI outfit IVO whose latest announcement in San Francisco begs the question of whether they have found something that the US$15 billion crowd have missed?

The Lawyer, the Engineer and the Venture Capitalist

Ivo was founded in 2021 by Jung, an Auckland Law School graduate who had also practised in Australia, and Jacob Duligall, a senior software engineer at Xero in Wellington (pictured above).

It is a very New Zealand origin story where he shared suspicion that contract review was the low-hanging fruit of an under-optimised profession.

It was not always called Ivo. The company began life as Latch and rebranded around its second seed round, led by Californian funds Uncork Capital and Fika Ventures.

The more consequential early cheque came from Daniel Gross, the investor who later co-founded Safe Superintelligence with Ilya Sutskever. Jung has said Gross did not just fund the business but talked him into relocating to San Francisco and, in the process, expanded the scope of his ambitions. The founders moved to the US in 2023.

Interestingly, Ivo did not sign its first New Zealand customer until after its Series A. It was built for the global market from day one, and it went where the AI talent and the AI money actually live.

The Money, in Brief

Ivo’s funding trajectory has been brisk rather than berserk:

• Seed rounds: including a US$4.8 million second seed led by Uncork and Fika.
• Series A (2025): US$16 million, led by Australian fund Blackbird.
• Series B (January 2026): US$55 million, again led by Blackbird, with Costanoa Ventures, Uncork, Fika, GD1 and Icehouse Ventures joining. A source told Reuters the round valued Ivo at around US$355 million.

That brings total funding to roughly US$75 million.

By way of comparison, Harvey closed a US$550 million round at a US$15.5 billion valuation last month. Stockholm’s Legora sits at US$5.6 billion and is reportedly chasing a figure closer to US$10 billion. Ivo could lose its entire valuation down the back of Harvey’s sofa and nobody at Harvey would notice.

Reuters reported that Ivo counts Uber, Shopify, IBM, Reddit and Canva among its users, with revenue up roughly sixfold in the year to January and headcount of 60 that it plans to triple by the end of 2026. Pinterest, Eventbrite and Fonterra have also been named as clients.

Where Ivo Fits in the Legal AI universe

Harvey and Legora are legal giants in the Legal AI world, while Ivo is a smaller but significant player in the legal AI universe.

Ivo is a smaller, denser planet. It does contracts, and it sells almost entirely to in-house teams rather than law firms. Its core product redlines agreements inside Microsoft Word against a company’s own negotiating playbook; its Intelligence product turns a company’s back catalogue of signed contracts into something you can actually query.

With this week’s general availability of Ivo Collaborate, it now claims to run the life of a contract from intake to signature and beyond.

That puts it in a crowded neighbourhood. As Artificial Lawyer observed when the Series B landed, in-house buyers now face general AI platforms with contract features bolted on, agentic workflow start-ups, a CLM industry that has caught up on generative AI, and the traditional ALSPs. Ironclad, Icertis, LegalOn and Spellbook all want the same general counsel’s budget.

Ivo’s pitch for standing out is less about raw intelligence and more about method. Jung told Reuters the company breaks contract review into more than 400 separate AI tasks rather than asking one model to do the whole thing.

His most memorable line for what most legal AI tools do is that they behave like “a smart junior lawyer who joined your company five minutes ago and has amnesia”.

So Why give it Away?

Ivo Sage is not a model built from scratch in a Kiwi garage. According to its Hugging Face model card, it is a set of fine-tuned weights layered on DeepSeek V4 Flash, the Chinese open-weights model, and trained with reinforcement learning on more than a thousand tasks from the Legal Agent Benchmark.

That benchmark was published by Harvey, so Ivo has taken its largest rival’s homework and used it to train a free model.

The results are genuinely interesting, if you read past the headline. On contract tasks, Sage lifts the base model’s criterion pass rate from 70 percent to 91 percent, the number in the press release.

But the stricter measure, where a task only counts if every single criterion is met, rises from 5.5 percent to 16.4 percent. The best frontier model in Ivo’s own comparison table manages 29.1 percent on that measure. The kicker is cost: roughly US$1.24 per contract task for Sage, against more than US$142 for that frontier model.

So this may not be the smartest model in the room, but a very competent one at a tiny fraction of the price, which is a perfectly sensible place to be, provided nobody mistakes 91 percent for “gets it right nine times out of ten”.

But there needs to be a word of caution for anyone imagining running this on the office laptop. The base model has 284 billion parameters, and Ivo’s own setup instructions assume a multi-GPU server.

“Download and run it for free” is true in the same sense that a Boeing is free to anyone with a runway. And some procurement teams will want to talk about the DeepSeek lineage, even though running the weights on your own hardware means your contracts go nowhere near China.

Legal AI company Equall released SaulLM-7B, an open legal language model under an MIT licence, back in 2024. Ivo’s claims it is the first open model post-trained for long, multi-step contract work. Lawyers, of all people, should appreciate a carefully drafted carve-out.

The Next Move

Jung’s argument is that the next leap in legal AI will not come from bigger models but from context such as the playbooks, the documents, the way a particular legal team actually works. Models, in this view, are becoming a commodity and so giving one away costs Ivo very little, earns it credibility with technical buyers and researchers, and quietly needles competitors whose valuations rest on proprietary magic.

Ivo’s headline finding is that frontier models are pushovers that they accept a competitor’s changes readily but push back badly, and they rarely escalate when they should. It is an unusually candid admission from a company that sells AI.

The Verdict

Ivo is not going to out-raise Harvey, and it is not trying to. What it has is a sharply defined niche, a blue-chip in-house client list, a founder who has done the job his software now does, and a willingness to publish research that does not flatter the technology.

Whether open-sourcing a model turns out to be a masterstroke or a footnote depends on whether Jung is right that context, not size, wins on the day.

If he is, the solicitor who got bored of supply agreements in Auckland may end up with the last laugh.