CHOOSEAgents & automation
Jev AI Model
Jeremy Jarry@JarryJeremy
Jev is an AI model designed to make decisions within software by returning choices, scores, or probabilities based on provided context and questions.
TakeGive it some context and questions and it returns choices, scores or probabilities.
Pattern✦ Decision layer primitives
One of the most talked-about new AI models on X can't write a sentence.
@typesafeai built Jev specifically to make decisions inside software. Give it some context and questions and it returns choices, scores or probabilities. It costs $0.042 per million input tokens, with no charge for output.
We've been looking at where this could fit inside our own agent fleet. Take one of our SEO agents. A lot of what it does isn't really "write something for me". It's decisions: classify a keyword, score an opportunity, decide whether a page has an issue, choose what should happen next or determine whether something needs human review.
I modelled 10,000 of those decisions using around 2,000 input tokens and 100 output tokens each. On Claude Sonnet 5, that's roughly $50 in model costs. With Jev handling the decision layer, the same 20M input tokens would cost around $0.84.
About 98% cheaper.
Obviously that doesn't make the whole SEO agent 98% cheaper. We still need larger models for research, reasoning and content generation. But it shows why I'm increasingly convinced that using one model for everything is the wrong architecture.
We've already found multi-model systems powerful because different models are better at different jobs. Jev pushes that further: some jobs don't need a language model designed to talk to humans at all.
At the scale of one agent, saving $49 isn't particularly exciting. At the scale of fleets making millions of decisions, it becomes a very different equation.
Related builds
