We've tried to take a first-principles approach to our end goal of 'legal superintelligence' that has involved identifying the areas of our domain most in need of improvement and releasing models that raise the bar on quality in those areas.

We've been around for a couple months now and ended up starting with retrieval and enrichment. The models we've released to tackle those problems have indeed been smaller in size than their competitors, yet they still rank ahead on open-source benchmarks.

Them being so small also helps with their accessibility — as I mention in our post, our models can be deployed on ordinary hardware, not a supercomputer.

Next on our roadmap is reasoning and research, which will require more infrastructure to support, but again, we aim to be judged by performance at the time of release.

> As much as I agree with the message, this reads like marketing copy trying to make a big deal out of a tiny model being hosted privately.

The point of this point is really just to reaffirm our commitment to sovereignty and accessibility and contrast our approach with that of major AI labs. It _is_ possible to commercialize LLMs while still keeping them accessible. A customer using a self-hosted deployment today does not need to worry about our models no longer being available tomorrow. We think that's a good thing. And moving forward, we want to keep that option available for anything we do, instead of trying to pull up the ladder while we're ahead.

Comments URL: https://news.ycombinator.com/item?id=48512915

Points: 98

# Comments: 21

Comments URL: https://news.ycombinator.com/item?id=48396657

Points: 2

# Comments: 0

My company makes legal AI models that we serve through an API. All our SDKs we're generated by Stainless. Those SDKs are in active use in production and, come September, will need to be maintained by someone other than Stainless.

We're not the only ones in this boat. OpenAI and Google also relied on Stainless.

What are other users planning on doing now that Stainless is gone?

We're hoping that a competitor or open-source community ends up taking on the mission of ensuring backwards compatibility or, at the very least, as seamless a migration as possible. If that doesn't get pulled off soon though, we may be forced into either completely breaking existing code or else maintaining everything by hand — the very thing Stainless was meant to take off our plate.

Has anyone else thought of better options yet?

Comments URL: https://news.ycombinator.com/item?id=48202774

Points: 5

# Comments: 3

There’s a massive issue with extrapolating to more complex tasks, however, where either you run the risk of prompt injection via granting your agent access to the internet or, more commonly, an exponential degradation in coherence over long contexts.

Comments URL: https://news.ycombinator.com/item?id=47485929

Points: 2

# Comments: 0

We make legal AI models. We recently released two state-of-the-art legal information retrieval models, Kanon 2 Embedder and Kanon 2 Reranker. Together, they rank first on Legal RAG Bench and the Massive Legal Embedding Benchmark (MLEB).

We also recently released Kanon 2 Enricher, an entirely new type of AI model capable of transforming lengthy, unstructured legal documents into highly structured knowledge graphs. We had the pleasure of having Harvey, KPMG Law, Alvarez & Marsal, Clifford Chance, Clyde & Co, Carey Olsen, Smokeball, and Moonlit be part of the closed beta for that model.

We're proud supporters of open source and maintain popular legal and AI datasets and libraries like the Open Australian Legal Corpus and the semchunk semantic chunking algorithm, together downloaded over one million times a month.

Our core mission is to solve every common data- and AI-related pain point of the legal tech industry. Uniquely, we're focused primarily on serving the legal tech industry rather than lawyers per se.

We, therefore, view folks like Harvey and Legora as customers instead of competitors.

In the few months since we've been around, we've witnessed massive growth in usage of our models by other new legal tech startups. We think our industry is only going to get bigger, particularly as more accessible legal services make their way into the hands of end consumers and enterprises instead of law firms.

We see ourselves as best placed to serve the needs of our industry given our strong experience and expertise in AI and law. I previously led all national-level AI projects at the Australian Attorney-General's Department as a senior data scientist while also holding an honors degree in law. My brother, in turn, is skilled in economics, data science, and AI.

We're currently working on scaling up our team to meet the growing demand we're seeing as well as help build out the next stage of our roadmap, which includes a first-of-a-kind knowledge graph of laws, decisions, and contracts from around the world, as well as the first legal reasoning model.

If you align with our mission and would like to partner with us or be part of our team, you can reach out at hello@isaacus.com.

Comments URL: https://news.ycombinator.com/item?id=47320176

Points: 1

# Comments: 0

We'd argue it makes a lot of sense to appropriate 'graphitization' as a term for a model designed to transform data into knowledge graphs.

Our pricing page can be found in our documentation here: https://docs.isaacus.com/pricing/prices. We're planning on making it more visible on our website; thanks for the feedback!

We do already have a government-facing client using one of our self-hosted deployments given the privacy and security concerns the legal industry tends to have (rightfully in our view) around AI.

In brief, our model is capable of: - Entity extraction, classification, and linking: identifying key entities like individuals, companies, governments, locations, dates, documents, and more, and classifying and linking them together. - Hierarchical segmentation: breaking a document up into its full hierarchy, including divisions, sections, subsections, paragraphs, and so on. - Text annotation: extracting common textual elements such as headings, sigantures, tables of contents, cross-references, and the like.

We built Kanon 2 Enricher from scratch. Every node, edge, and label in the Isaacus Legal Graph Schema (ILGS), which is the format it outputs to, corresponds to at least one task head in our model. In total, we built 58 different task heads jointly optimized with 70 different loss terms.

Thanks to its novel architecture, unlike your typical LLM, Kanon 2 Enricher doesn't generate extractions token by token (which introduces the possibility of hallucinations) but instead directly classifies all the tokens in a document in a single shot. This makes it really fast.

Because Kanon 2 Enricher's feature set is so wide, there are a myriad of applications it can be used for, from financial forensics and due diligence all the way to legal research.

One of the coolest applications we've seen so far is where a Canadian government built a knowledge graph out of thousands of federal and provincial laws in order to accelerate regulatory analysis. Another cool application is something we built ourselves, a 3D interactive map of Australian High Court cases since 1903, which you can find right at the start of our announcement.

Our model has already been in use for the past month, since we released it through a closed beta that included Harvey, KPMG, Clifford Chance, Clyde & Co, Alvarez & Marsal, Smokeball, and 96 other design partners. Their feedback was instrumental in improving Kanon 2 Enricher before its public release, and we're immensely thankful to each and every beta participant.

We're eager to see what other developers manage to build with our model now that its out publicly.

Comments URL: https://news.ycombinator.com/item?id=47229931

Points: 10

# Comments: 6

Comments URL: https://news.ycombinator.com/item?id=46863753

Points: 1

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=46793786

Points: 2

# Comments: 0

As @lopuhin points out, they already claimed that context window for previous iterations of GPT-5.

The funny thing is though, I'm on the business plan, and none of their models, not GPT-5, GPT-5.1, GPT-5.2, GPT-5.2 Extended Thinking, GPT-5.2 Pro, etc., can really handle inputs beyond ~50k tokens.

I know because, when working with a really long Python file (>5k LoCs), it often claims there is a bug because, somewhere close to the end of the file, it cuts off and reads as '...'.

Gemini 3 Pro, by contrast, can genuinely handle long contexts.

Comments URL: https://news.ycombinator.com/item?id=46086510

Points: 2

# Comments: 2

Comments URL: https://news.ycombinator.com/item?id=46013092

Points: 6

# Comments: 0

I'm excited to share the Massive Legal Embedding Benchmark (MLEB) — the first comprehensive benchmark for legal embedding models.

Unlike previous legal retrieval datasets, MLEB was created by someone with actual domain expertise (I have a law degree and previously led the AI team at the Attorney-General's Department of Australia).

I came up with MLEB while trying to train my own state-of-the-art legal embedding model. I found that there were no good benchmarks for legal information retrieval to evaluate my model on.

That led me down a months-long process working alongside my brother to identify or, in many cases, build our own high-quality legal evaluation sets.

The final product was 10 datasets spanning multiple jurisdictions (the US, UK, Australia, Singapore, and Ireland), document types (cases, laws, regulations, contracts, and textbooks), and problem types (retrieval, zero-shot classification, and QA), all of which have been vetted for quality, diversity, and utility.

For a model to do well at MLEB, it needs to have both extensive legal domain knowledge and strong legal reasoning skills. That is deliberate — given just how important high-quality embeddings are to legal RAG (particularly for reducing hallucinations), we wanted our benchmark to correlate as strongly as possible with real-world usefulness.

The dataset we are most proud of is called Australian Tax Guidance Retrieval. It pairs real-life tax questions posed by Australian taxpayers with relevant Australian Government guidance and policy documents.

We constructed the dataset by sourcing questions from the Australian Taxation Office's community forum, where Australian taxpayers ask accountants and ATO officials their tax questions.

We found that, in most cases, such questions can be answered by reference to government web pages that, for whatever reason, users were unable to find themselves. Accordingly, we manually went through a stratified sample of 112 challenging forum questions and extracted relevant portions of government guidance materials linked to by tax experts that we verified to be correct.

What makes the dataset so valuable is that, unlike the vast majority of legal information retrieval evaluation sets currently available, it consists of genuinely challenging real-world user-created questions, rather than artificially constructed queries that, at times, diverge considerably from the types of tasks embedding models are actually used for.

Australian Tax Guidance Retrieval is just one of several other evaluation sets that we painstakingly constructed ourselves simply because there weren't any other options.

We've contributed everything, including the code used to evaluate models on MLEB, back to the open-source community.

Our hope is that MLEB and the datasets within it will hold value long into the future so that others training legal information retrieval models won't have to detour into building their own "MTEB for law".

If you'd like to head straight to the leaderboard instead of reading our full announcement, you can find it here: https://isaacus.com/mleb

If you're interested in playing around with our model, which happens to be ranked first on MLEB as of 16 October 2025 at least, check out our docs: https://docs.isaacus.com/quickstart

Comments URL: https://news.ycombinator.com/item?id=45715623

Points: 11

# Comments: 0