Performance
Beyond the Frontier
Dynamically route every request to the best LLM in real-time. Get higher performance at lower costs while future-proofing your application.
Unlock exponential value as performance improvement and cost savings grow over time.
CUSTOM ROUTERS FOR ENTERPRISES
Optimize your performance and cost even further. Build custom routers specifically tailored for your needs.
01
BRING YOUR DATA
We collect your data — a sample of the types of prompts you will be sending to the router — to train an LLM Judge tailored to your requirements.
02
Choose your team of models
We use your custom Judge to assess the performance of different LLMs. This way, we'll identify which models adapt better to your requirements.
03
TRAIN YOUR ROUTER
With all the results, we will train your custom router so that it can predict the quality of responses of various models to new prompts. Based your quality constraints and the costs of each model, we can then efficiently route your requests.
04
USE AND REFINE
As you use the router, we collect performance data that allows us to refine model selection over time. The more you use it, the better it will get.
ROUTE RESPONSIBLY
Routing for Ethics and Safety
Most major models (and many lesser-known ones) have made a good effort towards model safety and ethical output, but the results across models is inconsistent — like other measures of quality, each model has strengths and weaknesses.
With effective routing, you can even out those inconsistencies in the ways that are important to you by including your specific ethical and safety concerns in your custom judge.
Ethics and safety become part of what “quality” means for your router.
With effective routing, you can even out those inconsistencies in the ways that are important to you by including your specific ethical and safety concerns in your custom judge.
Ethics and safety become part of what “quality” means for your router.
ROUTE RESPONSIBLY
Reduce Your Environmental Impact
Martian routers are able to predict the response quality of models given your prompts, which means we can route to smaller models that use fewer resources any time they will provide a sufficiently high-quality response.
Since smaller models that use fewer resources are almost always cheaper to run, routing aligns your economic interests with your environmental concerns.
Since smaller models that use fewer resources are almost always cheaper to run, routing aligns your economic interests with your environmental concerns.
