<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Models on yoav.blog</title><link>https://yoav.blog/tags/models/</link><description>Recent content in Models on yoav.blog</description><image><title>yoav.blog</title><url>https://yoav.blog/og-image.png</url><link>https://yoav.blog/og-image.png</link></image><generator>Hugo -- 0.155.1</generator><language>en-us</language><lastBuildDate>Thu, 22 Oct 2026 18:35:00 +0300</lastBuildDate><atom:link href="https://yoav.blog/tags/models/feed.xml" rel="self" type="application/rss+xml"/><item><title>Training small models for fun and profit (the easy way)</title><link>https://yoav.blog/2026/10/22/training-small-models-for-fun-and-profit/</link><pubDate>Thu, 22 Oct 2026 18:35:00 +0300</pubDate><guid>https://yoav.blog/2026/10/22/training-small-models-for-fun-and-profit/</guid><description>&lt;p&gt;Back in July 2026, Google announced they&amp;rsquo;ll &lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/model-versions"&gt;retire the Gemini 2.5&lt;/a&gt; family on October 20th.&lt;/p&gt;
&lt;p&gt;At &lt;a href="https://base44.com"&gt;Base44&lt;/a&gt; we had one last feature that was still using Gemini 2.5 Flash Lite: autocomplete in the builder, called millions of times a day. We had looked into a replacement model in the past, but nothing newer could match its latency. And for autocomplete, latency isn&amp;rsquo;t a nice-to-have: suggestions that arrive late don&amp;rsquo;t get seen, so higher latency = lower success rate.&lt;/p&gt;</description></item></channel></rss>