Model-opinionated
A multi-model architecture routes each task to the intelligence best suited to its accuracy, speed, reliability and cost requirements, without dependence on a single model or provider.
Model-opinionated
A multi-model architecture routes each task to the intelligence best suited to its accuracy, speed, reliability and cost requirements, without dependence on a single model or provider.
Model-opinionated
A multi-model architecture routes each task to the intelligence best suited to its accuracy, speed, reliability and cost requirements, without dependence on a single model or provider.
30+
Models and reasoning settings tested head-to-head
5
Validation stages before a model reaches production
3
Independent ways every model choice is proven
30+
Models and reasoning settings tested head-to-head
5
Validation stages before a model reaches production
3
Independent ways every model choice is proven
30+
Models and reasoning settings tested head-to-head
5
Validation stages before a model reaches production
3
Independent ways every model choice is proven
The model can change. The standard does not.
No single model does everything well
New AI models appear constantly, each with different strengths, weaknesses, speeds, costs and reliability.
Using just one model or allowing customers to choose their preferred provider avoids the question that users don’t realise matters most: Which model performs best for this specific task?
A model that excels at complex drafting may be unnecessarily slow for extracting contract data. A lightweight model that can process thousands of routine tasks quickly may not be appropriate for the hardest legal reasoning.
Using the biggest or newest model for everything does not guarantee the best result. It also creates unnecessary dependence on a single model or AI provider, increasing exposure to changes in availability, capacity, pricing or model deprecation.
The model can change. The standard does not.
No single model does everything well
New AI models appear constantly, each with different strengths, weaknesses, speeds, costs and reliability.
Using just one model or allowing customers to choose their preferred provider avoids the question that users don’t realise matters most: Which model performs best for this specific task?
A model that excels at complex drafting may be unnecessarily slow for extracting contract data. A lightweight model that can process thousands of routine tasks quickly may not be appropriate for the hardest legal reasoning.
Using the biggest or newest model for everything does not guarantee the best result. It also creates unnecessary dependence on a single model or AI provider, increasing exposure to changes in availability, capacity, pricing or model deprecation.
The model can change. The standard does not.
No single model does everything well
New AI models appear constantly, each with different strengths, weaknesses, speeds, costs and reliability.
Using just one model or allowing customers to choose their preferred provider avoids the question that users don’t realise matters most: Which model performs best for this specific task?
A model that excels at complex drafting may be unnecessarily slow for extracting contract data. A lightweight model that can process thousands of routine tasks quickly may not be appropriate for the hardest legal reasoning.
Using the biggest or newest model for everything does not guarantee the best result. It also creates unnecessary dependence on a single model or AI provider, increasing exposure to changes in availability, capacity, pricing or model deprecation.
THE LUMINANCE APPROACH
Luminance takes ownership of model selection.
Luminance has spent years building the evaluation systems, legal benchmarks and engineering know-how required to make these decisions systematically. Every new model enters the same process. When it delivers a stronger result for a specific task, it earns that role. When it does not, the existing model remains in production.
Luminance also develops proprietary specialist models. This means the model selected for a task can be purpose-built for legal work, as well as drawn from the wider model landscape.
This multi-model architecture is built for resilience. Work can be distributed across model families and providers, so no single model becomes a point of failure. It also supports independent validation, reducing the likelihood that the same failure mode appears across every model used to assess an output.
The result is a model strategy for every task that selects the strongest intelligence for each task while remaining adaptable as the AI landscape changes.
For every task in the platform, Luminance evaluates which model delivers the strongest combination of accuracy, speed, reliability and cost – in that order:
1) Accuracy: Does the model produce the best result for the legal task? Material improvements in accuracy take priority.
2) Speed: Where models perform to equivalent levels of accuracy, faster models allow contract analysis to happen more quickly and at greater scale.
3) Reliability: A model also must perform consistently within a live enterprise product. Availability and capacity matter alongside benchmark performance.
The result is not one preferred model. It’s a ranked model strategy for every task.
THE LUMINANCE APPROACH
Luminance takes ownership of model selection.
Luminance has spent years building the evaluation systems, legal benchmarks and engineering know-how required to make these decisions systematically. Every new model enters the same process. When it delivers a stronger result for a specific task, it earns that role. When it does not, the existing model remains in production.
Luminance also develops proprietary specialist models. This means the model selected for a task can be purpose-built for legal work, as well as drawn from the wider model landscape.
This multi-model architecture is built for resilience. Work can be distributed across model families and providers, so no single model becomes a point of failure. It also supports independent validation, reducing the likelihood that the same failure mode appears across every model used to assess an output.
The result is a model strategy for every task that selects the strongest intelligence for each task while remaining adaptable as the AI landscape changes.
For every task in the platform, Luminance evaluates which model delivers the strongest combination of accuracy, speed, reliability and cost – in that order:
1) Accuracy: Does the model produce the best result for the legal task? Material improvements in accuracy take priority.
2) Speed: Where models perform to equivalent levels of accuracy, faster models allow contract analysis to happen more quickly and at greater scale.
3) Reliability: A model also must perform consistently within a live enterprise product. Availability and capacity matter alongside benchmark performance.
The result is not one preferred model. It’s a ranked model strategy for every task.
THE LUMINANCE APPROACH
Luminance takes ownership of model selection.
Luminance has spent years building the evaluation systems, legal benchmarks and engineering know-how required to make these decisions systematically. Every new model enters the same process. When it delivers a stronger result for a specific task, it earns that role. When it does not, the existing model remains in production.
Luminance also develops proprietary specialist models. This means the model selected for a task can be purpose-built for legal work, as well as drawn from the wider model landscape.
This multi-model architecture is built for resilience. Work can be distributed across model families and providers, so no single model becomes a point of failure. It also supports independent validation, reducing the likelihood that the same failure mode appears across every model used to assess an output.
The result is a model strategy for every task that selects the strongest intelligence for each task while remaining adaptable as the AI landscape changes.
For every task in the platform, Luminance evaluates which model delivers the strongest combination of accuracy, speed, reliability and cost – in that order:
1) Accuracy: Does the model produce the best result for the legal task? Material improvements in accuracy take priority.
2) Speed: Where models perform to equivalent levels of accuracy, faster models allow contract analysis to happen more quickly and at greater scale.
3) Reliability: A model also must perform consistently within a live enterprise product. Availability and capacity matter alongside benchmark performance.
The result is not one preferred model. It’s a ranked model strategy for every task.
DIFFERENT TASKS REQUIRE DIFFERENT MODELS.
Match the model to the task
There is no single model running every Luminance workflow. Because models are tested by task, they are selected by task, too.
High-volume work
Routine, high-frequency work is routed to models that can deliver the required accuracy at much greater speed. This can include Luminance's own Luna models where specialist contract intelligence provides an advantage.
Complex legal reasoning
The hardest tasks are reserved for models that demonstrate the strongest reasoning capabilities, even where those models are slower or more expensive.
Extraction and classification
Models are selected according to the best measured balance of accuracy, speed and reliability for that specific job.
Large-scale review
When accuracy is equivalent, faster models can materially reduce the time required to process large document sets without sacrificing quality.
The result: The strongest model gets the job it has earned.
DIFFERENT TASKS REQUIRE DIFFERENT MODELS.
Match the model to the task
There is no single model running every Luminance workflow. Because models are tested by task, they are selected by task, too.
High-volume work
Routine, high-frequency work is routed to models that can deliver the required accuracy at much greater speed. This can include Luminance's own Luna models where specialist contract intelligence provides an advantage.
Complex legal reasoning
The hardest tasks are reserved for models that demonstrate the strongest reasoning capabilities, even where those models are slower or more expensive.
Extraction and classification
Models are selected according to the best measured balance of accuracy, speed and reliability for that specific job.
Large-scale review
When accuracy is equivalent, faster models can materially reduce the time required to process large document sets without sacrificing quality.
The result: The strongest model gets the job it has earned.
DIFFERENT TASKS REQUIRE DIFFERENT MODELS.
Match the model to the task
There is no single model running every Luminance workflow. Because models are tested by task, they are selected by task, too.
High-volume work
Routine, high-frequency work is routed to models that can deliver the required accuracy at much greater speed. This can include Luminance's own Luna models where specialist contract intelligence provides an advantage.
Complex legal reasoning
The hardest tasks are reserved for models that demonstrate the strongest reasoning capabilities, even where those models are slower or more expensive.
Extraction and classification
Models are selected according to the best measured balance of accuracy, speed and reliability for that specific job.
Large-scale review
When accuracy is equivalent, faster models can materially reduce the time required to process large document sets without sacrificing quality.
The result: The strongest model gets the job it has earned.
Model Evaluation
Every model earns its place
New models don’t automatically enter the Luminance Platform. Every candidate is evaluated against legal benchmarks, realistic contract work and expert judgment before it can replace the model already performing a task.
1. Survey the field
New models are assessed for their reported capabilities, strengths, and potential relevance to contract workflows.
2. Match models to tasks
Promising candidates are mapped to the specific tasks where their capabilities could provide an advantage.
3. Test individual tasks
The Luminance Test Harness benchmarks candidates on narrow, repeatable product tasks, measuring performance metrics, such as precision, recall, F-score, speed and reliability, against the model already in production.
4. Test the complete experience
Shortlisted models are evaluated through ALumNI, Luminance's end-to-end evaluation framework, using realistic user stories and legal tasks rather than isolated benchmark questions. It recreates realistic user stories and legal tasks across the complete workflow. Outputs are scored against task-specific rubrics built by Luminance’s in-house legal experts, while response speed is measured alongside quality.
5. Validate and release
The chosen model receives human sign-off before entering a phased production rollout that can be monitored and reversed.
Alongside this process, Luminance's legal experts independently blind-test promising models in the Luminance Arena, assessing qualities benchmark scores alone cannot capture, including tone, phrasing, verbosity and the overall quality of the legal output without knowing which model produced which answer.
A model must perform on the numbers and for the teams who will ultimately rely on it.
Model Evaluation
Every model earns its place
New models don’t automatically enter the Luminance Platform. Every candidate is evaluated against legal benchmarks, realistic contract work and expert judgment before it can replace the model already performing a task.
1. Survey the field
New models are assessed for their reported capabilities, strengths, and potential relevance to contract workflows.
2. Match models to tasks
Promising candidates are mapped to the specific tasks where their capabilities could provide an advantage.
3. Test individual tasks
The Luminance Test Harness benchmarks candidates on narrow, repeatable product tasks, measuring performance metrics, such as precision, recall, F-score, speed and reliability, against the model already in production.
4. Test the complete experience
Shortlisted models are evaluated through ALumNI, Luminance's end-to-end evaluation framework, using realistic user stories and legal tasks rather than isolated benchmark questions. It recreates realistic user stories and legal tasks across the complete workflow. Outputs are scored against task-specific rubrics built by Luminance’s in-house legal experts, while response speed is measured alongside quality.
5. Validate and release
The chosen model receives human sign-off before entering a phased production rollout that can be monitored and reversed.
Alongside this process, Luminance's legal experts independently blind-test promising models in the Luminance Arena, assessing qualities benchmark scores alone cannot capture, including tone, phrasing, verbosity and the overall quality of the legal output without knowing which model produced which answer.
A model must perform on the numbers and for the teams who will ultimately rely on it.
Model Evaluation
Every model earns its place
New models don’t automatically enter the Luminance Platform. Every candidate is evaluated against legal benchmarks, realistic contract work and expert judgment before it can replace the model already performing a task.
1. Survey the field
New models are assessed for their reported capabilities, strengths, and potential relevance to contract workflows.
2. Match models to tasks
Promising candidates are mapped to the specific tasks where their capabilities could provide an advantage.
3. Test individual tasks
The Luminance Test Harness benchmarks candidates on narrow, repeatable product tasks, measuring performance metrics, such as precision, recall, F-score, speed and reliability, against the model already in production.
4. Test the complete experience
Shortlisted models are evaluated through ALumNI, Luminance's end-to-end evaluation framework, using realistic user stories and legal tasks rather than isolated benchmark questions. It recreates realistic user stories and legal tasks across the complete workflow. Outputs are scored against task-specific rubrics built by Luminance’s in-house legal experts, while response speed is measured alongside quality.
5. Validate and release
The chosen model receives human sign-off before entering a phased production rollout that can be monitored and reversed.
Alongside this process, Luminance's legal experts independently blind-test promising models in the Luminance Arena, assessing qualities benchmark scores alone cannot capture, including tone, phrasing, verbosity and the overall quality of the legal output without knowing which model produced which answer.
A model must perform on the numbers and for the teams who will ultimately rely on it.
Explore the architecture
From the right context to the right answer.
Document Search Engine
Pre-computed legal knowledge and Context Graphs connect contracts, amendments and related documents, enabling users and AI agents to retrieve the right evidence faster and analyse transactions in context.
Differential Validation
Independent judges, grounded critics and task-specific checks scrutinise high-impact outputs against evidence, improving accuracy and making uncertainty visible when human review is required.
Luna Crescent
Luminance’s proprietary LLM is purpose-built for specialist legal tasks, improving the speed and economics of high-volume contract extraction, classification and analysis.
Explore the architecture
From the right context to the right answer.
Document Search Engine / Context Graphs
Pre-computed legal knowledge and Context Graphs connect contracts, amendments and related documents, enabling users and AI agents to retrieve the right evidence faster and analyse transactions in context.
Differential Validation
Independent judges, grounded critics and task-specific checks scrutinise high-impact outputs against evidence, improving accuracy and making uncertainty visible when human review is required.
Luna Crescent
Luminance’s proprietary LLM is purpose-built for specialist legal tasks, improving the speed and economics of high-volume contract extraction, classification and analysis.
Explore the architecture
From the right context to the right answer.
Document Search Engine / Context Graphs
Pre-computed legal knowledge and Context Graphs connect contracts, amendments and related documents, enabling users and AI agents to retrieve the right evidence faster and analyse transactions in context.
Differential Validation
Independent judges, grounded critics and task-specific checks scrutinise high-impact outputs against evidence, improving accuracy and making uncertainty visible when human review is required.
Luna Crescent
Luminance’s proprietary LLM is purpose-built for specialist legal tasks, improving the speed and economics of high-volume contract extraction, classification and analysis.
GET STARTED
Get more from your contracts
See how Luminance can help your teams negotiate smarter, surface what matters, and keep contracts moving across the enterprise.
GET STARTED
Get more from your contracts
See how Luminance can help your teams negotiate smarter, surface what matters, and keep contracts moving across the enterprise.
GET STARTED
Get more from your contracts
See how Luminance can help your teams negotiate smarter, surface what matters, and keep contracts moving across the enterprise.
© 2026 Luminance. All rights reserved.