Litil / Blog

Blog

Model releases, benchmark results, and notes from our work on legal AI.

Earlier posts

Research note

What legaltech companies are filing patents on

Six companies, their patent filings, and what we think the application histories tell us.

6 min read

Model training

Soup du Jour

Train the same model three times. Average the weights. Here’s how model soup took our clause classifier from 77.8% to 92.2%.

5 min read

Research paper

Models can all agree on legal risk, but wildly diverge on what to do about it.

Six model families often caught the same legal risks. They did not make the same trade-offs—and the contract they started from shaped the deal they produced.

20 min read

Release note

LiTiL models for organizing contracts and extracting their terms

Clause Classifier and Contract Extractor work alongside our legal search models to turn contract passages into usable information.

7 min read

Models / benchmarks

Introducing LiTiL’s models for legal search

LiTiL released three specialized models for legal query-to-passage retrieval.

6 min read

Release note

Ten specialist models for legal work

Meet ten LiTiL releases for intake, privacy, contract analysis, and specialist review.

3 min read

Release note

The LiTiL model family: ten small models for a legal intelligence stack

Ten open, specialized models with defined jobs across intake, routing, privacy, contract analysis, and specialist review.

6 min read

Model note

Route the request before doing the expensive work: LiTiL Legal Request Router 1.5B

A small model that turns an incoming legal request into a structured route before the system loads the next workflow.

3 min read

Model note

LiTiL Legal Intake 4B: bounded decisions at the front of a legal workflow

A bounded-output model for intake decisions, routing codes, issue codes, citations, and short extracted values.

3 min read

Model note

A privacy checkpoint for legal documents: LiTiL Legal PII Finder

A legal-document privacy checkpoint that returns labeled sensitive spans with source offsets.

3 min read

Model note

LiTiL ClauseCheck 1.5B: turn contract text into a provision map

A small contract model that checks whether a requested provision appears and helps build a reviewable provision map.

4 min read

Model note

LiTiL ClauseTagger 4B: give contract passages a common vocabulary

A multi-label contract model that gives passages a consistent taxonomy for indexing and downstream routing.

3 min read

Model note

Give every contract clause a stable name: LiTiL Clause Classifier

A contract classifier that assigns one category to a provision and returns scores across the available taxonomy.

3 min read

Model note

LiTiL Contract Extractor 1.7B: capture the term and keep the source

A contract extraction model that returns the language answering a specific question or a stable absence value.

3 min read

Model note

LiTiL Contract Playbook 3B: apply the rule and return the next action

A contract-review model that applies a supplied playbook rule and returns the next structured review action.

4 min read

Model note

A reviewable Howey worksheet from transaction evidence: LiTiL-Howey-2b

A structured Howey-test model that maps numbered transaction evidence to the four factors for review.

4 min read

Model note

Compare the deal term to the draft: LiTiL Financing Review 9B

A focused financing-review model that compares an agreed term with the corresponding draft provision.

4 min read

Field note

The Murky Middle

The obvious cases are easy to label. Legal judgment lives in the part that depends on context.

2 min read

Research note

Train for the Outlier

The judgment worth teaching is not always the answer most people chose.

2 min read

Position

Where the Model Fits

A model should have a defined job in the stack, and the organization should be able to replace it.

2 min read

Field note

A Correct Answer Can Still Fail the Assignment

Legal work ends in something another person has to review, use, send, sign, or file. That changes how agents should be built and measured.

5 min read

Systems note

The Model Learns the Grader

A reward puts pressure on the model. A weak grader can turn its own mistakes into training signal.

5 min read

Position

Open Means You Can Leave

Ownership matters most when the model changes, the contract ends, or the vendor disappears.

4 min read