MiniMax-M2.7 legal REAP releases
Formal model cards and release pages for the 0.35 and 0.45 pruning lanes, plus GGUF artifacts for local legal AI evaluation on owned hardware.
PLI Labs — the open arm
Open weights, carved for legal work.
PLI Labs publishes expert-pruned model checkpoints optimized for the practice of law, with open weights, plus open-source tools for legal document work. Download them, inspect them, run them on your own machines — no account, no API, no meter.
5+
public model repositories
MiniMax-M2.7 legal REAP checkpoints and GGUF artifacts published for local evaluation on owned hardware.
Prunes
not finetunes
No client data, no training pass — structural prunes of open checkpoints, so confidentiality questions never arise.
V100 & B70
serving recipes published
Reproducible launch profiles for accessible, older GPUs — the recipes almost nobody else documents.
The method
Prunes, not finetunes.
The strongest open AI models today are mixture-of-experts models: inside one large model live many specialist sub-networks (“experts”), and only a few activate for any given sentence. Run such a model on a steady diet of legal work — pleadings, discovery, financial records, case analysis — and a clear pattern emerges: some experts do the heavy lifting, and many barely fire at all.
What is a legal prune?
A prune measures which experts a legal workload actually uses, then removes the ones it doesn’t. What remains is a smaller, faster checkpoint of the same model that preserves its legal capability while fitting on far more modest hardware. The weights that survive are the original weights — nothing is retrained on new data, which is why a prune is not a finetune. It is the same craftsman’s toolbox with the tools you never reach for taken out of the bag.
Every Labs release ships with a model card describing the base model, the pruning recipe, the evaluation method, and known limitations — enough detail to reproduce or challenge our results.
Releases
Checkpoints & repositories
Formal model cards, Hugging Face release pages, GGUF artifacts, and owned-hardware launch profiles for B70 and V100 inference lanes.
Qwen3.5 legal REAP releases
Two Qwen3.5 legal-compression candidates published for lawyer-supervised evaluation, with a candid finding: the model is prune-resistant, so PLI Labs published a conservative expert cut rather than force a bigger compression headline.
Why we publish
The profession needs machines it can audit.
Lawyers are asked to trust AI systems they cannot inspect, hosted in places they cannot see, under terms that change without notice. We think the long-term answer for the profession is the opposite: open weights a firm can download, examine, and run inside its own walls — with the duty of confidentiality satisfied by architecture rather than by promises.
Publishing our prunes and tools openly also keeps us honest. Our claims about what works for legal AI are testable by anyone with a graphics card, and our consulting clients benefit from methods that have survived public scrutiny.
The two arms feed each other: lessons from consulting engagements and a working litigation practice tell Labs what to build, and Labs gives firms a no-cost way to evaluate PLI’s work before hiring anyone.