Ten specialist models for legal work
Meet ten LiTiL releases for intake, privacy, contract analysis, and specialist review.
LiTiL Labs has released ten specialist models for specific jobs in legal workflows, from routing a request to extracting a contract term and applying a playbook.
A request needs a route. A document may need sensitive information marked. Someone needs to find the clause, pull out the term, and compare it with the company's rule. Our Open Intelligence Models collection covers those jobs with ten public model repositories.
What the models do
| Model | Useful for |
|---|---|
| Legal Request Router 1.5B | Sorting incoming work into the right legal queue |
| Legal Intake 4B | Returning a short action, code, or value from a fixed set of choices |
| Legal PII Finder | Marking personal and sensitive information in legal text |
| ClauseCheck 1.5B | Checking whether a named contract provision is present |
| ClauseTagger 4B | Attaching several contract labels to one passage |
| Clause Classifier | Giving a provision one category from a fixed taxonomy |
| Contract Extractor 1.7B | Pulling out the contract language that answers a specific question |
| Contract Playbook 3B | Applying a supplied playbook rule and returning the next review action |
| LiTiL-Howey-2b | Connecting supplied transaction facts to the four Howey factors |
| Financing Review 9B | Comparing an agreed financing term with the draft provision |
Each model has one bounded job. One general model does not need to make every decision in a legal workflow.
How the pieces work together
In a contract workflow, the router can select the right process. The PII model can mark sensitive text. A retrieval model can find the likely clauses. The extractor can pull out the term, and the playbook model can compare it with the company's rule.
Each step returns something the application can inspect. The source document and the company's instructions stay beside the model output. A person can see what the system found, what rule it used, and where the next decision came from.
This is what we mean by LiTiL's open intelligence layer. It is a set of specialist components between legal text and the product interface. A team can use one model without adopting the whole collection. It can also replace one part without rebuilding the rest of the system.
Put a model to work
Start with a task your team already repeats. Use the loading example in the model card, keep the expected output narrow, and test it against your own documents before it enters a live workflow.
The collection is on Hugging Face. Each model card describes the input, output, loading steps, limits, and evaluation.
Credits and licenses
The model cards record the base-model lineage and training sources for each release. The collection builds on work from Qwen, Fastino and GLiNER2, Answer.AI and ModernBERT, and the authors and maintainers of LegalBench, LexGLUE and LEDGAR, and CUAD, among others. We thank and credit those teams.
Built with Qwen. Contract Playbook 3B uses Qwen2.5-3B-Instruct under the Qwen Research License. That license limits use to noncommercial research and evaluation unless the user has a separate commercial license.
The repositories do not all use the same license. As of September 14, 2026, five are labeled Apache-2.0 and five are labeled other. Use the terms shown on the specific model page, and keep any required base-model and dataset notices when you redistribute it. A public repository is not, by itself, permission to ignore the upstream terms.