<?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: Kamil ArndtKamil Arndt</title>
    <description>The latest articles on DEV Community by Kamil ArndtKamil Arndt (@kamil_arndt_f23b17c940b24).</description>
    <link>https://dev.to/kamil_arndt_f23b17c940b24</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.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3903304%2Feb891231-0fa1-4d59-816a-1183255c4dbd.png</url>
      <title>DEV Community: Kamil ArndtKamil Arndt</title>
      <link>https://dev.to/kamil_arndt_f23b17c940b24</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/kamil_arndt_f23b17c940b24"/>
    <language>en</language>
    <item>
      <title>I switched from OpenAI to z.ai for codiai coding review ng and I'm genuinely happy with it — honest review</title>
      <dc:creator>Kamil ArndtKamil Arndt</dc:creator>
      <pubDate>Wed, 29 Apr 2026 00:49:43 +0000</pubDate>
      <link>https://dev.to/kamil_arndt_f23b17c940b24/i-switched-from-openai-to-zai-for-codiai-coding-review-ng-and-im-genuinely-happy-with-it-honest-1p2m</link>
      <guid>https://dev.to/kamil_arndt_f23b17c940b24/i-switched-from-openai-to-zai-for-codiai-coding-review-ng-and-im-genuinely-happy-with-it-honest-1p2m</guid>
      <description>&lt;p&gt;I've been using z.ai for about three months now. Here's what I've learned.&lt;/p&gt;

&lt;h2&gt;
  
  
  What z.ai is
&lt;/h2&gt;

&lt;p&gt;z.ai (by Zhipu AI) offers a subscription-based API for their GLM models. The selling point: flat monthly pricing instead of per-token billing. You pay 8/mo (Lite), 2/mo (Pro), or 60/mo (Max), and you get API access to their coding models.&lt;/p&gt;

&lt;p&gt;The current flagship is GLM-5.1. It's the #1 open-source model and #3 globally on LMArena Code, reportedly on par with Claude Opus 4.6.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why I signed up
&lt;/h2&gt;

&lt;p&gt;Two reasons: cost and tool sprawl.&lt;/p&gt;

&lt;p&gt;On cost - I was burning through OpenAI and Anthropic API credits faster than I'd like to admit. Productive weekend? 0-100 gone. Sometimes more.&lt;/p&gt;

&lt;p&gt;On tools - I use Claude Code for reviews, Cline for generation, Aider for CLI work, and occasionally others. Each wanted its own API key. Each had its own billing. It was a mess.&lt;/p&gt;

&lt;p&gt;z.ai gave me one key that works with all of them. And the bill is the same every month regardless of how much I use it (within quotas).&lt;/p&gt;

&lt;h2&gt;
  
  
  What's good
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Tool support is real.&lt;/strong&gt; Claude Code, Cline, Cursor, Windsurf, Aider, Continue.dev, Kilo Code - they all work. You point them at z.ai's endpoint and go. I've tested maybe 8-9 of these personally.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The model is solid for most tasks.&lt;/strong&gt; Code generation, test writing, debugging, documentation, refactoring small to medium chunks - GLM-5.1 handles these well. Not perfect, but good enough that I reach for it without thinking.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;MCP tools included.&lt;/strong&gt; Vision Analysis, Web Search, Web Reader, Zread. These are part of the plan, not extras.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Pricing is predictable.&lt;/strong&gt; No surprises. No checking my API usage dashboard nervously. I know what I'll pay this month because it's the same as last month.&lt;/p&gt;

&lt;h2&gt;
  
  
  What's not good
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;It's not Claude.&lt;/strong&gt; GLM-5.1 is good, but for nuanced code understanding - figuring out what a developer was trying to do in a messy codebase, suggesting just the right abstraction - Claude is still better. Noticeably.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Response times vary.&lt;/strong&gt; Sometimes fast, sometimes slow. Peak hours are worse. If you're used to the snappiness of a local model or a well-provisioned API, the inconsistency can be frustrating.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Hallucinations happen.&lt;/strong&gt; More often than with Claude, less often than with GPT-3.5. The model will sometimes confidently generate code that looks right but does the wrong thing. You need to review everything.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Context window limitations.&lt;/strong&gt; In long sessions, GLM-5.1 loses the thread. I've had it forget what file we were working on, or suggest changes that contradict something we agreed on 20 minutes earlier. Restarting the session fixes it, but it breaks flow.&lt;/p&gt;

&lt;h2&gt;
  
  
  The pricing breakdown
&lt;/h2&gt;

&lt;p&gt;For context, here's what I was paying before vs. now:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Before (OpenAI API)&lt;/th&gt;
&lt;th&gt;Now (z.ai Pro)&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Monthly cost&lt;/td&gt;
&lt;td&gt;00-200 (variable)&lt;/td&gt;
&lt;td&gt;2 (fixed)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Tools supported&lt;/td&gt;
&lt;td&gt;Configure each separately&lt;/td&gt;
&lt;td&gt;One key, ~20 tools&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Bill predictability&lt;/td&gt;
&lt;td&gt;Nope&lt;/td&gt;
&lt;td&gt;Yep&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;For someone on a budget or just tired of variable API bills, the math is straightforward.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who this is for
&lt;/h2&gt;

&lt;p&gt;If you code for a living and use AI tools daily, z.ai is worth trying. The Lite plan at 8/mo is low enough risk that you can test it without commitment.&lt;/p&gt;

&lt;p&gt;If you're a casual user who fires up Copilot once a week, you probably don't need this.&lt;/p&gt;

&lt;p&gt;If you need Claude-level quality and budget isn't a concern, stick with Anthropic. z.ai is good but it's not that good.&lt;/p&gt;

&lt;h2&gt;
  
  
  My setup
&lt;/h2&gt;

&lt;p&gt;All three point at z.ai's API endpoint. One key.&lt;/p&gt;

&lt;p&gt;Claude Code - reviews, refactoring, documentation&lt;br&gt;
Cline (VS Code) - generation, debugging&lt;br&gt;&lt;br&gt;
Aider - CLI work, quick fixes&lt;/p&gt;

&lt;h2&gt;
  
  
  Setup (5 minutes)
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;Sign up at z.ai&lt;/li&gt;
&lt;li&gt;Pick a plan (Lite starts at 8/mo)&lt;/li&gt;
&lt;li&gt;Get your API key&lt;/li&gt;
&lt;li&gt;Configure your tools to use &lt;a href="https://api.z.ai/v1" rel="noopener noreferrer"&gt;https://api.z.ai/v1&lt;/a&gt; as the endpoint&lt;/li&gt;
&lt;li&gt;Done&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Bottom line
&lt;/h2&gt;

&lt;p&gt;z.ai is the best value I've found for AI coding. Not the best model - that's still Claude or GPT-4 depending on the task. But the best value. The combination of flat pricing, broad tool support, and a model that's good enough for daily work is hard to beat.&lt;/p&gt;




&lt;p&gt;Disclosure: If you sign up through my referral link, you get 10% off and I get credits. I genuinely use this tool daily - the referral is a bonus, not the reason I wrote this. Here it is: &lt;a href="https://z.ai/subscribe?ic=LXVCVV38ZL" rel="noopener noreferrer"&gt;https://z.ai/subscribe?ic=LXVCVV38ZL&lt;/a&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>coding</category>
      <category>llm</category>
      <category>productivity</category>
    </item>
  </channel>
</rss>
