<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Sam</title>
    <description>The latest articles on DEV Community by Sam (@shofficial10).</description>
    <link>https://dev.to/shofficial10</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4254672%2F93831534-7f64-4746-8c9c-26e3e582a3db.png</url>
      <title>DEV Community: Sam</title>
      <link>https://dev.to/shofficial10</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/shofficial10"/>
    <language>en</language>
    <item>
      <title>Your AI model is retiring. Here's what actually breaks.</title>
      <dc:creator>Sam</dc:creator>
      <pubDate>Fri, 20 Nov 2026 01:36:57 +0000</pubDate>
      <link>https://dev.to/shofficial10/your-ai-model-is-retiring-heres-what-actually-breaks-2g3m</link>
      <guid>https://dev.to/shofficial10/your-ai-model-is-retiring-heres-what-actually-breaks-2g3m</guid>
      <description>&lt;h1&gt;
  
  
  Your AI model is retiring. Here's what actually breaks.
&lt;/h1&gt;

&lt;p&gt;Two big deadlines are close. Anthropic retires &lt;code&gt;claude-sonnet-4-5-20250929&lt;/code&gt; on &lt;strong&gt;Nov 30, 2026&lt;/strong&gt;, and requests to it will fail after that. OpenAI shuts down &lt;code&gt;gpt-5&lt;/code&gt;, &lt;code&gt;gpt-5-mini&lt;/code&gt;, &lt;code&gt;gpt-5-nano&lt;/code&gt;, &lt;code&gt;gpt-5-pro&lt;/code&gt;, &lt;code&gt;o3&lt;/code&gt; and &lt;code&gt;o3-pro&lt;/code&gt; on &lt;strong&gt;Dec 11, 2026&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;This month I sent 20 small pull requests to open-source apps still using models like these, and 8 have been merged so far. The lesson from all of them: changing the model name is rarely the whole job. Here's what else breaks, and the fix for each.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Find every place the model is named
&lt;/h2&gt;

&lt;p&gt;Search the whole repo, not just the API call: code, environment variables, config files, model dropdowns, tests, docs and examples. Search for both forms of the name, like &lt;code&gt;gpt-5-mini&lt;/code&gt; and &lt;code&gt;gpt-5-mini-2025-08-07&lt;/code&gt;, or &lt;code&gt;claude-sonnet-4-5&lt;/code&gt; and &lt;code&gt;claude-sonnet-4-5-20250929&lt;/code&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Claude Sonnet 4.5 to Sonnet 5.5
&lt;/h2&gt;

&lt;p&gt;Anthropic's replacement is &lt;code&gt;claude-sonnet-5-5&lt;/code&gt;. Its migration guide lists several changes that turn working requests into errors:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Sampling settings are rejected.&lt;/strong&gt; Remove &lt;code&gt;temperature&lt;/code&gt;, &lt;code&gt;top_p&lt;/code&gt; and &lt;code&gt;top_k&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Thinking is on by default.&lt;/strong&gt; A response can now start with a thinking block, so code that reads &lt;code&gt;content[0].text&lt;/code&gt; breaks. Read content blocks by their &lt;code&gt;type&lt;/code&gt;. This is the bug I fixed in &lt;a href="https://github.com/GuyMannDude/mnemo-cortex/pull/25" rel="noopener noreferrer"&gt;mnemo-cortex #25&lt;/a&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Thinking budgets are rejected.&lt;/strong&gt; Replace &lt;code&gt;budget_tokens&lt;/code&gt; with &lt;code&gt;thinking: {"type": "adaptive"}&lt;/code&gt; and an &lt;code&gt;output_config.effort&lt;/code&gt; level.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The old way to turn thinking off fails.&lt;/strong&gt; &lt;code&gt;{"type": "disabled"}&lt;/code&gt; returns a 400 error; send &lt;code&gt;{"type": "between_tools"}&lt;/code&gt; instead.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Prefill is rejected.&lt;/strong&gt; The conversation must end with a user message, so move any prefilled assistant text into the system prompt.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Forced tool use is rejected.&lt;/strong&gt; A &lt;code&gt;tool_choice&lt;/code&gt; of &lt;code&gt;any&lt;/code&gt; or &lt;code&gt;tool&lt;/code&gt; returns a 400 error; use &lt;code&gt;auto&lt;/code&gt; with strict tools.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The same text uses about 30% more tokens.&lt;/strong&gt; Recheck &lt;code&gt;max_tokens&lt;/code&gt;, limits and cost.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  3. OpenAI's gpt-5 family
&lt;/h2&gt;

&lt;p&gt;OpenAI's deprecations page lists these replacements:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Shutting down Dec 11&lt;/th&gt;
&lt;th&gt;Recommended replacement&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;gpt-5&lt;/code&gt;, &lt;code&gt;o3&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;&lt;code&gt;gpt-6.1-sol&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;
&lt;code&gt;gpt-5-pro&lt;/code&gt;, &lt;code&gt;o3-pro&lt;/code&gt;
&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;gpt-6.1-sol&lt;/code&gt; with &lt;code&gt;reasoning.mode: pro&lt;/code&gt; (Responses API)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;gpt-5-mini&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;&lt;code&gt;gpt-6.1-terra&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;code&gt;gpt-5-nano&lt;/code&gt;&lt;/td&gt;
&lt;td&gt;&lt;code&gt;gpt-6-luna&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;What to test after you switch:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Tool calls.&lt;/strong&gt; When I moved a Telegram bot from &lt;code&gt;gpt-5-mini&lt;/code&gt; to &lt;code&gt;gpt-5.6-terra&lt;/code&gt;, the maintainer added one more commit before merging: "disable Terra reasoning when using function tools" (&lt;a href="https://github.com/kirill-markin/chatgpt-telegram-bot-telegraf/pull/43" rel="noopener noreferrer"&gt;PR #43&lt;/a&gt;). If your app uses tools, run those paths before you ship.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reasoning settings.&lt;/strong&gt; Check that your reasoning effort value is one the new model accepts. GPT-5.6 Terra, for example, doesn't list &lt;code&gt;minimal&lt;/code&gt;.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cost.&lt;/strong&gt; Compare the new model's price per token with what you pay today before you switch everything.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  4. Already happened elsewhere
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Groq&lt;/strong&gt; shut down &lt;code&gt;llama-3.3-70b-versatile&lt;/code&gt; and &lt;code&gt;llama-3.1-8b-instant&lt;/code&gt; on Aug 16; calls to them now return errors.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;xAI&lt;/strong&gt; redirects eight retired Grok model names, mostly to &lt;code&gt;grok-4.3&lt;/code&gt;, which bills at its own rate.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mistral&lt;/strong&gt; returns a 404 error for retired model IDs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Google&lt;/strong&gt; lists Nov 17 as the earliest shutdown date for several Gemini preview audio models.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Checklist
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Search the whole repo for every form of the old model name.&lt;/li&gt;
&lt;li&gt;Remove settings the new model rejects.&lt;/li&gt;
&lt;li&gt;Read responses by block type, not by position.&lt;/li&gt;
&lt;li&gt;Test tool calls and reasoning settings.&lt;/li&gt;
&lt;li&gt;Recheck token limits and cost.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;If you maintain an open-source app still on one of these models, tell me and I'll send a free PR. If you'd rather have it done in your private code, I do it for a fixed price, from $400 per repo, paid after you've reviewed it: [your email], or github.com/SamHofmannOfficial.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;I used AI tools to help research and write this guide.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>llm</category>
      <category>openai</category>
      <category>softwaredevelopment</category>
    </item>
  </channel>
</rss>
