---
format: "aidr-story-markdown/v1"
id: "8324d57fee88899a02d3293eea1f7666a3a44c5d596c12e9bc73589696d3d691"
canonical_url: "https://aidr.today/8324d57f?lang=en"
title: "Training a 3.8B LLM to 0.384 CORE for $998 – Hugo Vergnes"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-10T02:04:11.000Z"
category: "Research"
topics: ["llm","training","inference","open-source","fine-tuning"]
source_urls: ["https://hugovergnes.github.io/little-lm-3-8b/","https://news.ycombinator.com/item?id=49637435"]
summary: "A 3.8B model pretrained from scratch to 0.384 CORE on 65B tokens in 43 hours, for $998."
---

# Training a 3\.8B LLM to 0\.384 CORE for $998 – Hugo Vergnes

> [Open the canonical story](<https://aidr.today/8324d57f?lang=en>)

**Published:** 2026-09-10T02:04:11.000Z
**Category:** Research
**Topics:** llm, training, inference, open\-source, fine\-tuning

## Summary

A 3\.8B model pretrained from scratch to 0\.384 CORE on 65B tokens in 43 hours, for $998\.

## Sources

- [Story source](<https://hugovergnes.github.io/little-lm-3-8b/>)
- [Discussion](<https://news.ycombinator.com/item?id=49637435>)