<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Marin — USASI news</title>
    <link>https://unitedstatesofamericasuperintelligence.com/open/marin/</link>
    <atom:link href="https://unitedstatesofamericasuperintelligence.com/open/marin/feed.xml" rel="self" type="application/rss+xml"/>
    <description>Dated, sourced news items in the USASI catalog that mention Marin. Independent project. Not a United States government website.</description>
    <language>en-us</language>
    <lastBuildDate>Tue, 13 Oct 2026 12:00:00 GMT</lastBuildDate>
    <item>
      <title>Marin begins pretraining Marin 535B-A23B, a mixture-of-experts model</title>
      <link>https://unitedstatesofamericasuperintelligence.com/news/marin-begins-pretraining-535b-a23b-model/</link>
      <guid isPermaLink="true">https://unitedstatesofamericasuperintelligence.com/news/marin-begins-pretraining-535b-a23b-model/</guid>
      <pubDate>Tue, 13 Oct 2026 12:00:00 GMT</pubDate>
      <category>research</category>
      <description>In a launch note dated September 3, 2026, first published on the Open Athena blog and mirrored by Marin, Marin said it had started pretraining Marin 535B-A23B a couple of weeks earlier. The note describes a mixture-of-experts model with 535 billion total parameters and 23 billion active, planned to pretrain for about three months on 18 trillion tokens, and says the team ported its infrastructure and training stack from Google TPUs on GCP to NVIDIA GPUs on CoreWeave before launch, with post-training to follow. It says the entire run, including architecture, data, and predicted loss, is documented on GitHub; it does not give a date or license for releasing the model's weights.</description>
    </item>
  </channel>
</rss>
