The Jev Question

In September 2026 TypeSafe AI launched Jev, which it calls a “System One” model. Jev is a decision model, so you give it a question and a set of options, and it returns a fast, structured answer. It doesn’t chat, and it doesn’t explain itself.

This sets Jev apart from models like ChatGPT, which generate answers one token at a time, requiring a full pass through the model for every token. Jev reads the question once and returns its decision in a single pass.

Named after a Manchester economist

Interestingly, the name “Jev” is inspired by William Stanley Jevons. In 1865, this Manchester economist (whose office at Owens College, then on Quay Street, was just a few minutes’ walk from Fuzzy Labs!) published The Coal Question.

The most famous passage is about efficiency. James Watt’s steam engine got far more work out of each ton of coal than the engines before it, so you might expect Britain to have burned less. Instead, it burned more. Cheaper power meant more engines, more mills, and more railways.

Jevons meant his paradox as a warning that making a resource cheaper to use makes it run out faster. TypeSafe borrows his name to reference the same effect, betting that every drop in the cost of intelligence will unlock dramatically more uses for it.

And if you name an efficiency technology “Jev”, and the patron saint of efficiency backfiring did his thinking in Manchester, you should expect us at Fuzzy Labs to raise an eyebrow.

The price of intelligence

On price, Jev is cheap. The independent JevBench benchmark puts it at about $0.05 per thousand correct decisions, five to seven times less than comparable models from OpenAI and Google.

The frontier models are getting cheaper too. OpenAI and Anthropic both cut their prices in September, and over a longer view, Epoch AI found the price of a given level of AI performance falling about 40× a year.

However, to follow Jevons’ argument more faithfully, we need to look past the price of intelligence and at the physical resource behind it. That gives us a modern version of his question, how much energy should a decision cost?

One model for every job

For much of the last decade, the recipe for improving AI has looked remarkably simple. Much like the steam age, more power is achieved with more fuel: give general-purpose models more computation, and let them scale. The result is that large language models have become something close to a universal tool for AI.

A customer asking “where is my parcel?” can go through the same machinery that drafts contracts and develops mathematical proofs. For jobs like that, a frontier language model can be vastly more complex than the task requires, a bit like firing up a mill engine to boil a kettle.

Energy per decision

I wanted to see how Jev’s efficiency claims compared with conventional language models on energy, so I benchmarked Jev against a range of other systems on the same 231 decisions and estimated the energy each used per correct decision. Those estimates are approximate, particularly for closed models where I had to work back from price, so they should read as a comparison rather than exact measurements.

Here’s what came out:

• Jev got 200 of 231 right, at about 30 joules per correct decision (roughly what a 10-watt LED bulb uses in three seconds).
• The big general-purpose models got 225 or 226 right, but used roughly 150–450 joules per correct decision.
• Between models giving nearly the same answers, energy per correct decision varied as much as fifteenfold.

Going from Jev’s 200 correct to the big models’ 225 is nearly all about the hardest tasks: long policy documents, multi-hop database lookups, and detailed arithmetic.

So the practical question is which decisions really need a frontier model. When I re-scored the results by sending the hard task types to GPT-5.6 Luna and everything else to Jev, the combination got 97.0% of decisions right against 97.4% for Luna alone, at under half the cost.

Tokens are getting cheaper, and we’re using far more of them

Google processed 9.7 trillion tokens a month in May 2024 and 3.2 quadrillion in May 2026, which is roughly a 330× increase in two years. The International Energy Agency expects data-centre electricity use to more than double by 2030, to slightly more than Japan uses today, and AI is the main driver.

Multiply a fifteenfold difference in energy by that kind of growth, and the choice of engine starts to matter at massive scale.

Choosing the right engine

None of this means big models are the wrong choice. They really are better at the hard problems. The question is whether every decision needs one.

For each job, it’s worth asking what the decision actually needs. Some need a frontier model. Many can be handled by something smaller, faster and cheaper, often a model you can run and control yourself. Getting that mix right is a design choice, and it’s the kind of choice we help our clients make.

What do we burn the coal for?

Jevons was right about coal. Britain’s coal production kept climbing for almost fifty years after he wrote The Coal Question, eventually peaking in 1913. The industrial revolution transformed everyday life, driving extraordinary improvements in living standards. But the same energy revolution also drove the climate change we are struggling to tackle today.

TypeSafe is betting that Jevons will be right about intelligence too, and the numbers suggest that cheaper intelligence will make far more of it economically viable. The important question is what we choose to do with it.

So the question from Manchester is no longer simply how long the fuel will last. It’s which engines we use, and what we use them for.

For the full benchmark, the charts and how I estimated the energy, read the complete article on our MLOps.WTF substack.

‍