Answer typed yes-or-no questions about text in 33 milliseconds

Made by Convai Innovations

A small model built to make decisions rather than write text.

You hand it a piece of text — an email, a support ticket, a JSON record — together with a set of questions and the allowed answers, and it returns the answers with a probability attached, all in one pass of about 33 milliseconds across more than 100 languages. It never generates prose, so there is no output to parse and nothing to hallucinate. It is 421 million parameters: a ModernBERT-large backbone, a bidirectional text encoder, with a decision head trained on top. The creator reports batched throughput of 103 to 332 questions a second on a Tesla T4.

You could use it to…

Route a support ticket before the page finishes loading
Answer yes-or-no questions from a fixed set of choices
Score a batch of tickets across a hundred languages at once

Can I use this?

Needs a GPU or CPU

You'll need A GPU or a reasonable CPU, and Python · Setup needed

Worth knowing It answers only from the options you give it, so it cannot tell you something you did not think to ask, and its probabilities are only as meaningful as the calibration behind them. Every benchmark figure is the creator’s own; the comparison against a commercial alternative uses that product’s published numbers on different prompts and sample sizes, which the creator states openly. Loading a different language checkpoint costs seconds, so mixed-language traffic needs the router configured to keep more than one in memory. Apache 2.0. BuildTube has not run or verified this model.

Why it's here

Jev's training method (RLCD) applied to 33-millisecond typed decisions in 100+ languages.

Part of “The Jev wave”, a BuildTube editorial story

Among the top trending AI models on Hugging Face

4,753 people have liked it on Hugging Face.
It was downloaded 0 times in the last 30 days.

Numbers from the snapshot taken 1 October 2026; not refreshed since.

Behind it

See the code on Hugging Face

BuildTube has not run or verified this project. Everything above is written from what the creator published.

Made this?