<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
  <title>MATRIX · Timestral</title>
  <link>https://derekwang2002.github.io/ai-intelligence/en/projects/timestral/</link>
  <description>Material changes in AI models, agents, engineering and infrastructure, with primary sources and recommendations.</description>
  <language>en</language>
  <lastBuildDate>Mon, 02 Nov 2026 16:56:00 GMT</lastBuildDate>
  <atom:link href="https://derekwang2002.github.io/ai-intelligence/en/feeds/projects/timestral.xml" rel="self" type="application/rss+xml"/>
  <item>
    <title>Mistral releases four open-weight Timestral forecasting models</title>
    <link>https://derekwang2002.github.io/ai-intelligence/en/events/ev-20261015-01/</link>
    <guid isPermaLink="false">ev-20261015-01-published</guid>
    <pubDate>Thu, 15 Oct 2026 16:00:25 GMT</pubDate>
    <category>New</category>
    <category>TRIAL</category>
    <description>&lt;p&gt;Mistral released four downloadable Timestral time-series models at 5M, 28M, 123M and 209M parameters. One checkpoint handles univariate and multivariate probabilistic forecasts with past and known-future covariates. The public 209M card lists 32,768 time steps of context, a 1,024-step forecast horizon and 21 quantiles. Its GIFT-Eval, TIME and fev-bench rankings are vendor-submitted and not independently reproduced.&lt;/p&gt;&lt;p&gt;&lt;strong&gt;Why read: &lt;/strong&gt;Timestral lowers the trial cost of time-series foundation models with 5M–209M open weights, while leaderboard advantages still need independent reproduction.&lt;/p&gt;&lt;p&gt;Recommendation: TRIAL&lt;/p&gt;&lt;p&gt;Sources: &lt;a href=&quot;https://mistral.ai/news/timestral/&quot;&gt;Introducing Timestral&lt;/a&gt; · &lt;a href=&quot;https://huggingface.co/mistralai/Timestral-209M&quot;&gt;Timestral-209M model card&lt;/a&gt;&lt;/p&gt;</description>
  </item>
</channel>
</rss>
