GLM 5 (Lite Coding Plan) + Skills + Pi : This $3 Setup BEATS Claude Code & Codex!!!
Key Takeaways
The video discusses the GLM 5 model and its capabilities, particularly with the Lite Coding Plan, and how it can be used with various tools such as Kilo CLI, Pi, and Claude Code, to achieve similar results to more expensive models like Claude Code Max or Codex, for a significantly lower cost, around $10 a month or $84 a year.
Full Transcript
[music] >> Hi. Welcome to another video. So, GLM-5 is now available in the light coding plan of ZAI. Which means that you can basically get one of the most advanced models for just $10 a month or $27 per quarter or $84 per year. Which is like 60% cheaper than a month of Claude Code Max or a Codex subscription. Which is really insane. Remember that it is still being rolled out on the light plans. So, you might not see it immediately. But, it should be there soon. Anyway, there's also the pro plan that costs $30 a month or $252 a year. Which is also really cheap. Considering how good GLM-5 is. I mean, it's really on par with Opus. And if you combine it with good skills and stuff, then it will be a really very good and cheap option to use. So, I wanted to tell you guys how you can use it as well. I'll be mainly focusing on two things. Which are how to configure it and how to get better results out of it by using some of the skills and tips as well. I have been using the GLM coding plan for a while now. So, I have some good observations about it. But, before we get into that, I want to quickly talk about today's sponsor On Demand. If you've been trying to build AI agents, you know the pain. You find one tool for summarization, another for web scraping, another for data analysis. And then you spend more time stitching them together than actually getting work done. On Demand solves this. On Demand is a platform where you can discover, assemble, and automate AI agents all in one place. They have an agent marketplace with over 400 agentic tools ready to go. So, instead of hunting for tools and figuring out APIs, you just browse, pick what you need, and deploy. That gives you over 1,200 possible agent combinations, which is insane. But, the cool part is the playground. You take these agents and assemble them into a workflow for your exact use case. And once it works, you turn it into a full automation using their flow builder. It's visual, no code, and lets you chain multi-agent workflows that run repeatedly without you touching anything. You can also bring your own models, and it's privacy first with secure connectors. Sign up through my link in the description, and you'll get $5 in free credit. Go check it out. First of all, to set it up, you can set it up almost anywhere. Like, you can connect it in open code, Pi, Kilo, and even Claude Code. I use it with Kilo CLI more these days. You can connect it quite easily there. You can first install Kilo CLI I quite easily with a simple command. And then you can just run it. And then you'll need to connect the ZAI coding plan. To do that, you can run the slash connect command and then select the ZAI coding plan option. And then enter the API key that you get from the console platform. And then just put it there. Once that's done, you can just run the slash models command and select the GLM-5 model and just start using it right away. This works quite well. And you can also install skills and everything. Many people also like using the Pi coding agent, and it is much more customizable. So, you can get even better results. To use it, you can get Pi quickly installed. And then you can just set up the API key for it and get it running. So, that should also be cool. One thing that you can do to get better results with GLM-5 is that you can install the front-end design skill by Anthropic. It really improves things when you're trying to build a front-end with it as well. So, that is something to consider. Most of the skills that you see in Anthropic's skill library work really well with GLM as well. So, you can also just install the GLM coding plan in Claude Code and easily toggle skills as well if you want. Or just download them and then put them in Kilo CLI or something. I generally put these in Kilo CLI and let it have them. And it works quite well. Also, if you do want a graphical interface rather than the CLI, then what you can do is go to VS Code and install Kilo Code there. And then just configure the GLM coding plan by going to the settings. And then you can use it quite easily. You can also use it with something like Conductor by just configuring Claude Code with the GLM coding plan. And then Conductor uses the Claude agent SDK that directs it to the GLM coding plan models. And you can then get good results with it as well by having multiple agents work in conjunction and so on. Also, GLM coding plan is an always thinking model. So, you can't like adjust the reasoning effort that much. But, you can actually ask it to think less. And it thinks less. Which is pretty good. And many times, it overthinks. Which can sometimes be good. But, mostly it results in worse generations. So, I'd recommend that you just ask in your prompt for it to think less. And it does adhere to that mostly. Also, as the context length increases over 200,000, I'd recommend that you start a new thread and then work with that. Because context rotting is a much bigger issue with this model than with something like Claude. It is mostly apparent in their coding plans. So, I'd recommend that you create new threads more often for the optimal and best results. Also, this is not a vision model. So, if you're on the pro plan, then you get web search and an MCP tool for vision that allows GLM to send your screenshots to a vision model. And then it can give back the text content to GLM-5. And it can then work accordingly. Another way to use this is with the Z Code UI. Which is actually by ZAI themselves. So, you can potentially download this and it should mostly work out of the box and allow you to use your coding plan with Claude Code. But, also give you a simple and easy-to-use GUI over it as well. Which is kind of cool if you ask me. It is really great as well. And you will probably get a bit more flexibility in performance as this is their own thing. I am still surprised that they haven't made their own agentic contraption yet. It's really interesting. So, that is also an option. I think GLM-5 with the coding plan gives you the same capabilities for just way less money. Which is really insane. I mean, it costs you like 10% of what you'll pay for Claude or Codex and gives you like 90% of the same capabilities. I also know that there is some great stuff coming from them. Which I can't wait to talk about. But, yeah. They are cooking. And this plan will be getting even more worth it in the coming days. Which is great. So, that is about it. I really think that if you're a student or someone who likes to budget and save money, then this might be a really, really good option for you. You can use it with things like open Claude as well. And it is much easier to get return on this investment than with something like Claude or Codex. So, yeah. Just go ahead and save some money with this plan. You get the MCP tools with it. And you also get the Z code that you can use as well. And you are getting all this for like $80 a year. Which is insane value if you ask me. Like, the price to performance is just on another level. Overall, it's pretty cool. Anyway, share your thoughts below and subscribe to the channel. >> [music] >> You can also donate via super thanks option or join the channel as well and get some perks. I'll see you in the next [music] video. Bye.
Original Description
Visit OnDemand: https://app.on-demand.io/auth/signup?refCode=AICODEKING_MI1
In this video, I'll be sharing my first impressions of Pony Alpha 2, a new model from Z AI that appears to be fine-tuned for OpenClaw. I compare it with GLM 5 in coding and agentic workflows, talk about its speed, long-context retention, and writing quality, and explain why it feels like a better fit for day-to-day OpenClaw usage.
--
Resources:
GLM-5 Coding Plan (affiliate link - not sponsored): https://z.ai/subscribe?ic=NWKPDIY9WD
--
Key Takeaways:
🚀 Pony Alpha 2 feels extremely fast compared to GLM 5 and responds almost instantly in many workflows.
🧠 The model seems better optimized for agentic tasks, tool calling, and instruction following inside OpenClaw.
💻 While it is decent at coding, it does not seem to outperform GLM 5 on coding-heavy prompts.
🛠️ Pony Alpha 2 works especially well with Skills, such as presentation workflows and web crawling tasks.
📚 Long-context retention appears stronger, with better history handling and smarter tool reuse when needed.
✍️ Writing quality also seems slightly improved, especially for research-oriented workflows.
🌍 If it launches with open weights and good pricing, it could become a very interesting model for everyday AI agents.
Watch on YouTube ↗
(saves to browser)
Sign in to unlock AI tutor explanation · ⚡30
More on: LLM Engineering
View skill →Related Reads
📰
📰
📰
📰
Bitemporal AI Memory: How to Preserve What an Agent Knew Then
Dev.to · Ethan Beirne
I run my one-person business with 12 AI employees. Here's the actual org chart.
Dev.to · Luna
Why Agentic AI Needs More Than Standard Model Risk Management
Medium · AI
What is Anyscale? The Platform Powering Scalable AI and Python Applications
Medium · Startup
🎓
Tutor Explanation
DeepCamp AI