The new Judgement API is a capability that’s powered under the hood by the new Jev model by TypeSafe AI. What it does is provide an extremely fast way to classify information, and at a small fraction of the cost.
What is it good for, then?
There are many workflows where the task we use AI for is “what is this information like”, “who should handle this”, or many similar things. This is where this model and capability shines — the Judgement API can take any text input, and given a number of options will choose which is the most likely one.
Say you want to classify a customer support ticket to figure out which team should be handling it. That’s exactly what this API is amazing for. Partly because it’s blazing fast, and because it will cost considerably less to use an LLM for it.
Judgement API in action
Each incoming ticket goes in with four teams as options. The Judgement API picks the one most likely to own it, and the ticket is routed there.
#48210maria@northwind.co
Charged twice this month
Our card was charged twice for the September invoice. Can you refund the duplicate payment?
Which team should handle this ticket?
#48210 · Charged twice this month
For example, if your team handles 200,000 customer tickets a month and you want to analyze first which team should handle them. Assuming 275 tokens per ticket (roughly equivalent to a long paragraph), that’s 55 million tokens, and on Anthropic’s Claude Sonnet 5, that would cost anywhere from $110 to $180 to process.
Using the Judgement API, this would cost about $30.
You can start using the Judgement API in your workspace right now, and if you’ve already built something that classifies content, you can ask Luo to update that to use the new API to reduce your token usage:
Ask Luo
“Please analyze the spec of this workspace and see if there are opportunities to use the platform’s new Judgement API to make processing faster and cheaper.”
As always, we’d love feedback, so if you end up using the new capability, let us know what you think!