Luminance launches Luna Crescent contract AI model

Luminance has launched Luna Crescent, a proprietary artificial intelligence model for contract work. It says the model outperformed general-purpose large language models in contract analysis tests.

Luna Crescent is the first model in Luminance's Luna series, designed specifically for legal and commercial contract tasks, and is already embedded across its contract intelligence features.

According to Luminance, the model was trained on the structure and language of legal agreements rather than broader consumer or business text. That approach, it says, improves precision in tasks such as identifying liability caps, termination rights and confidentiality obligations.

To assess the model, Luminance's research and development team in Cambridge built an internal benchmark called ContractIQ Bench. The test suite includes more than 180,000 manually annotated and reviewed legal data points drawn from contract concepts and provisions.

Using that benchmark, Luminance compared Luna Crescent with leading general-purpose models on contract-understanding tasks involving previously unseen legal concepts. According to the company, the specialist model delivered 5% higher accuracy and generated answers up to four times faster.

Luminance also says legal experts preferred Luna Crescent's outputs in blind evaluations. Those assessments focused on identifying complex contractual concepts, extracting key information and interpreting nuanced legal language.

Benchmark tests

Luminance says the model performed better in areas that often challenge broader language models, including spotting non-standard clause constructions, detecting concepts expressed in unfamiliar wording, classifying obligations buried in dense boilerplate and identifying what is missing as well as what is present in a contract.

According to Luminance, Luna Crescent processes text at roughly 200 to 400 tokens per second. Based on those figures, work that might take 30 seconds on a general-purpose model could be completed in under eight seconds.

That speed matters because contract analysis often spans large document collections across legal, procurement, finance and operations teams. Faster review can affect how quickly businesses surface risks, check obligations and extract commercial information from contract portfolios.

Model strategy

Luminance says Luna Crescent was developed within its own artificial intelligence stack and fine-tuned on part of a proprietary legal dataset drawn from more than 220 million verified legal documents. It says this gives customers more visibility into how its legal AI systems are built, governed and operated.

The model also sits within the platform's wider multi-model setup. That structure allows Luminance to apply different models, or combinations of models, to legal tasks across the contract lifecycle, including review, negotiation and post-signature management.

Luminance was founded by AI researchers from the University of Cambridge and focuses on software for enterprise contracting. The company says its platform is used by more than 1,000 large organisations in more than 70 countries.

The announcement reflects a broader shift in artificial intelligence software toward narrower models trained for specific professional tasks rather than general-purpose use. In legal technology, that trend has focused on whether specialist training can improve reliability in areas where wording, structure and context carry significant legal and commercial consequences.

In contract work, the distinction matters because many tasks depend on recognising not just keywords but the relationship between rights, obligations, restrictions and risk allocation. Contract review systems are increasingly expected to identify unusual drafting, compare terms across documents and flag the absence of clauses as well as their inclusion.

Graham Sills, Co-Founder and Director of AI at Luminance, said: "With Luna Crescent, enterprises finally have an LLM built for their work rather than borrowed from elsewhere. This is what vertical AI should deliver: greater accuracy, greater speed and greater trust for the work that matters most. By developing Luna Crescent within our own AI stack, we can give customers clarity over how their contract intelligence is built, governed and applied to their business, and power legal-grade agents capable of taking precise, context-aware action across the contract lifecycle."