Jev, GPT, or both?

Jev is a small, fast, very cheap model that answers a typed question and says how sure it is. The hybrid lets Jev answer alone when it is sure, and hands the rest to a big LLM. Every answer below was recorded once per item; move the slider and the three setups are recomputed from those real answers.

1 · Pick a task

2 · Set the rule

How sure must Jev be to answer alone? Below that, the item goes to the LLM.

Jev alwaysLLM always90%

What does one wrong answer cost you? Decides the winner.

Loading…

3 · All tasks at a glance (same slider, same price of a mistake)

Jev onlyHybridLLM only★ = winner at this slider and mistake price

4 · The items themselves

Text	Right answer	Jev says · how sure	Goes to	LLM says	Hybrid

5 · Try your own task

Runs Jev (typesafe/jev-1.13) and an LLM of your choice live from this page through OpenRouter, on your own examples. Your key stays in this browser tab (optionally in its local storage) and is only ever sent to openrouter.ai. You see a cost estimate before anything is called.

Your OpenRouter key

remember it in this browser · get a key

LLM (any OpenRouter model id) Possible answers (comma-separated) The question (one line)

Examples, one per line — add => answer when you know the right answer

How the two are asked

Jev gets a typed choice question: your question as the instruction, one criterion per answer (the answer itself, unless you write answer: criterion in the list). The LLM gets the same question and list in a prompt and must reply with JSON {"label": …}. Prices are OpenRouter's public list; the real cost shown after the run is what OpenRouter reports per call.