---
format: "aidr-story-markdown/v1"
id: "712dc859868826bbc21f3eb7dc1ca76c13444ae60d15b32fe18238d2890a996b"
canonical_url: "https://aidr.today/712dc859?lang=en"
title: "Marin Starts Training 535B AI Model on 18.75T Tokens in Biggest Open Run"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-08-21T23:49:27.000Z"
category: "Research"
topics: ["llm","training","chips"]
source_urls: ["https://huggingnews.com/ai/marin-starts-training-535b-ai-model-on-1875t-tokens-in-biggest-open-run-ff328597"]
summary: "The Marin team began training its 535B-A23B large language model this week using 11 GB200 NVL72 clusters. The process involves pretraining on 80% and midtraini…"
---

# Marin Starts Training 535B AI Model on 18\.75T Tokens in Biggest Open Run

> [Open the canonical story](<https://aidr.today/712dc859?lang=en>)

**Published:** 2026-08-21T23:49:27.000Z
**Category:** Research
**Topics:** llm, training, chips

## Summary

The Marin team began training its 535B\-A23B large language model this week using 11 GB200 NVL72 clusters\. The process involves pretraining on 80% and midtraini…

## Sources

- [Story source](<https://huggingnews.com/ai/marin-starts-training-535b-ai-model-on-1875t-tokens-in-biggest-open-run-ff328597>)