# MotherDuck adds prompt_jev() for SQL classification

> MotherDuck says its new SQL function classified 100,000 rows in 40 seconds for $0.50, versus 31 minutes 59 seconds and $37.58 for a frontier LLM comparison.

- Publication: Go Big News
- By Nifal Adam, Go Big News Staff (Staff News)
- Category: AI
- Published: 2026-09-22T20:31:18.531+00:00
- Canonical: https://www.gobignews.com/ai/motherduck-adds-prompt-jev-for-sql-classification

---

[MotherDuck](https://motherduck.com) has added prompt_jev(), a SQL function powered by TypeSafe’s Jev model, for text classification in its analytics platform. The company said in a post on X that the function makes classification about 50 times faster at about 1% of the cost.

On a 100,000-row benchmark, Jev completed the task in 40 seconds for $0.50, MotherDuck said. The frontier LLM comparison took 31 minutes 59 seconds and cost $37.58, according to MotherDuck’s blog post.

The test used 100,000 articles from the AG News training split for four-class news-topic classification. MotherDuck reported 89% accuracy for Jev and 88% for the comparison model, identified in the blog post as gpt-5.6-terra. It reported throughput of 2,484 rows a second for Jev, compared with 52 rows a second for the LLM.

The function takes a text column, an instruction and classification choices, then returns a label, score and confidence for each result. MotherDuck said prompt_jev() is available on all paid plans and is billed through its standard billing.

[TypeSafe AI](https://typesafe.ai/blog/introducing-system-one-models-and-jev) calls Jev a System One Model. TypeSafe says the model accepts unstructured state and returns typed probabilistic decisions, using parallel, non-autoregressive sampling and its Reinforcement Learning for Calibrated Decisions training approach. The company says Jev runs 40 to 200 times faster than existing frontier models on comparable tasks. TypeSafe was founded by Diogo Almeida, who previously worked at OpenAI on methods for making language models follow instructions, and spent two years in stealth before its announcement.

MotherDuck, founded in 2022 by CEO Jordan Tigani, Ryan Boyd and Tino Tereshko, builds a serverless cloud analytics platform on the open-source DuckDB database. The Seattle company raised a $52.5 million Series B in September 2023 after a combined $47.5 million seed and Series A. It later raised a $33 million Series B+ in May 2025, bringing total funding to roughly $133 million, according to the cited sources.

---

## Companies in this story

- Motherduck: https://www.gobignews.com/company/motherduck
- Typesafe: https://www.gobignews.com/company/typesafeai

---

## X profiles in this story

- @motherduck: https://www.gobignews.com/profile/motherduck