There is something quite strange about how powerful AI has become.
You can give Claude a complicated problem, watch it figure things out, build something useful, and do work that would have taken you hours. It still feels a little crazy how far we’ve come.
But sometimes, even after the difficult part is finished, Claude keeps working. It goes through the same files again, checks little things, changes its mind, and uses more tokens along the way.
I think there is a much bigger opportunity hiding in this than people realize.
And with two recent AI releases, we’re getting a pretty interesting way to take advantage of it.
A Very Interesting Combination Has Arrived
On September 22, Anthropic released Claude Opus 5.5, one of its most powerful models for coding, reasoning, and building AI agents.
Then, on October 14, TypeSafe released Jev 1.14, an improved version of a rather unusual AI model.
The interesting part is what happens when you put these two together.
Opus can handle the difficult thinking, create software, solve problems, and work through complicated projects. Jev can make the smaller decisions that happen around that work.
And Jev is incredibly cheap.
At current API prices, one million input tokens costs $4 with Opus 5.5 and just $0.042 with Jev. That makes Jev roughly 95 times cheaper for input tokens alone.
Of course, that doesn’t mean your entire AI workflow becomes 95 times cheaper. The actual savings depend on how much unnecessary work you can remove.
But a 10x or even 20x reduction in certain workflows is a realistic target worth testing.
I find this particularly interesting for people using Claude Code, Cowork-style tools, coding agents, automated workflows, and AI systems that need to keep working through many steps.
Because once these models are given the right jobs, you can build some seriously useful things.
Let’s get into the setup.
Read more
