---
format: "aidr-story-markdown/v1"
id: "4b99a853d3afdbe8b64f2b036fbd7294e109d34e0dac21f17b0aa5801a8bf391"
canonical_url: "https://aidr.today/4b99a853?lang=en"
title: "Training a 4B model to produce 81% faster query plans than Postgres"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-16T18:50:00.000Z"
category: "Research"
topics: ["qwen","llm","agent","rl","database","inference"]
source_urls: ["https://rohanbansal.com/qorl","https://news.ycombinator.com/item?id=49731285"]
summary: "...or how to make Qwen learn query optimization via agentic reinforcement learning"
---

# Training a 4B model to produce 81% faster query plans than Postgres

> [Open the canonical story](<https://aidr.today/4b99a853?lang=en>)

**Published:** 2026-09-16T18:50:00.000Z
**Category:** Research
**Topics:** qwen, llm, agent, rl, database, inference

## Summary

\.\.\.or how to make Qwen learn query optimization via agentic reinforcement learning

## Sources

- [Story source](<https://rohanbansal.com/qorl>)
- [Discussion](<https://news.ycombinator.com/item?id=49731285>)