Transcription
[music] >> Hi. Welcome to another video.
So, this video will be more of a comparison video. And by comparison, I don't mean that this is going to be a super scientific benchmark breakdown where I pull up 20 charts and pretend that one number explains everything. That is not really how AI coding tools work anymore. The model matters, of course. Benchmarks matter. Raw coding ability matters, but the plan also matters. The limits matter. The tooling matters. The ecosystem matters. The way you are allowed to use the model matters. And honestly, the feeling of whether you can code freely or whether you're constantly watching your usage meter matters a lot as well.
So, in this video, I want to talk about the four coding subscriptions that I think are the most interesting right now. We have GLM coding plan. We have Claude Code. We have Codex. And we have Kimmy. These four are not exactly the same kind of product, but they are fighting for the same type of user. They are all trying to be the thing you pay for when you want an AI coding workflow that can actually help you build, refactor, debug, review, and ship real projects.
And that is why this comparison is interesting because if you are a casual user, almost anything is fine. You can use the free tier of one tool, the trial of another tool, and just jump around. But if you are someone who codes every day, then the question becomes different. It is not just which one is the smartest. It is which one gives me enough usage to actually work. Which one can handle front end and back end. Which one is flexible enough to use in the tools I already like. Which one feels reliable when the code base gets messy. And which one is worth paying for every month. That is the real comparison.
So, let's start with GLM coding plan. GLM coding plan is interesting because it gives you access to GLM 5.1 and GLM 5.1 is a very good model. It is strong at front end. It is good at back end. It can reason through a larger coding task without instantly falling apart. It can make decent UI decisions. It can understand existing code. And it can usually follow project structure better than a lot of cheaper models. It is not perfect, obviously. No coding model is perfect, but GLM 5.1 feels like one of those models that can actually sit in the serious coding category instead of the fun demo model category.
The thing that makes the GLM coding plan different is not just the model, though. It is the flexibility. You are not locked into one official app. You can use it with tools like Claude Code, Kilocode, Cursor, OpenCode, Kline, and other coding agents that support this kind of provider setup. And that is a huge deal. Because a lot of these coding subscriptions are basically saying, "Pay us every month, but only use the model inside our own product." GLM is more open than that. If you like Kilocode, you can use it there. If you like Claude Code's workflow, you can route it there. If you want to test it in another agent, you can do that as well. That alone makes the plan feel more useful because it lets you choose the workflow instead of forcing the workflow on you. I use it with Kilocode myself, especially the Kilo CLI. The setup is very easy. You run the connect command, select GLM coding plan, enter your API key, and then you can start using it. That kind of setup is exactly what I like because it does not make the subscription feel trapped.
Now, the big downside is pricing. GLM increased the pricing by a lot. The plan is now around $18, $72, and $160 a month depending on the tier you choose. The pricing does go down if you take a yearly or quarterly plan, but still, it is more expensive than it used to be. So, I don't want to pretend that GLM is just this tiny cheap option anymore. It is not. It has moved closer to the serious paid coding plan category. But even with that, the value can still be strong because the flexibility is strong. If you are only using one official app, then maybe the pricing feels harder to justify. But if you are the kind of person who jumps between Kilocode, Claude Code style workflows, Cursor, Kline, and other tools, GLM coding plan becomes much more interesting. It becomes less like paying for one app and more like paying for a coding model that can move with you. That is the best argument for GLM. It gives you GLM 5.1. And it lets you use it in the places where you actually want to code.
Next, let's talk about Claude Code. Claude Code is still one of the most important AI coding tools. I don't think anyone can deny that. It changed how a lot of people think about terminal coding agents. It made the workflow feel natural. It gave people a way to talk to an agent inside a repo and let it inspect files, edit code, run commands, and iterate. And the models are still very good. When Claude is good, it is really good. For front end work, Claude has often felt stronger than GPT models. It tends to understand design better. It can produce nicer UI. It can make cleaner visual choices. And it usually has a better sense of what a usable front end should look like. For back end work, it is also capable. It can read code bases well. It can reason about bugs. And it can explain changes clearly.
So, the problem with Claude Code is not that the product is bad. The problem is the value equation. The $20 plan can be useful, but it is low if you are doing serious daily coding. The $100 and $200 plans are more serious, but those prices are not easy to recommend anymore. And the biggest issue is that you are mostly tied to Claude Code itself. You are paying for the Claude Code is still premium, but it is premium in a way that now feels harder to justify for a lot of people. If you are an Anthropic power user and you love Claude Code and you can afford the higher plans, then sure, it can still make sense. But if you're asking me what I would recommend to most people, I don't think Claude Code is the automatic answer anymore. It used to be easier to say, "Just use Claude Code." Now, there are too many strong alternatives.
One of those alternatives is Codex. Codex has been getting better in a very noticeable way. And what makes Codex interesting is that it is not just one thing. You have the local coding workflow. You have the ChatGPT side. You have the app experience. You have cloud tasks. You have code review features. You have the broader OpenAI ecosystem around it. That makes Codex feel like more of a complete coding platform rather than just a terminal tool. The pricing also makes it interesting. Codex starts from the free tier, which already gives people a way to try it. Then the $20 plan gives you a lot more room. And if you move to the $100 or $200 plans, the limits become much bigger. OpenAI also keeps improving how the usage resets and how Codex fits into the rest of the subscription. So, for a lot of people, the $20 Codex option is going to feel much more reasonable than jumping straight into a $100 or $200 dedicated coding subscription.
The model itself is really good, too. Codex is strong at understanding code bases. It is good at refactors. It is good at explaining what it is doing. It is good at long coding sessions where you want it to keep track of the project instead of treating every prompt like a fresh one-off task. It is also very good when you use it for back end work, debugging, tests, scripts, architecture, and general engineering tasks. The only real weakness I still see is front end taste. Sometimes Codex can make front end UIs that are technically correct, but visually bland. It can build the thing, but the design might not feel as sharp as Claude or GLM. It can overuse generic layouts, generic spacing, generic cards, and generic colors if you do not guide it properly. But that weakness is fixable. If you use good instructions, skills, design rules, examples, or a strong front end workflow, Codex improves a lot. That is why I still rate Codex very highly. Because the baseline engineering ability is strong. The ecosystem is strong. The limits are competitive. And the tool keeps improving. If you already use ChatGPT for research, writing, image work, voice, planning, or general AI tasks, Codex becomes even easier to justify. You are not only paying for a coding agent, you're paying for a broader AI subscription where the coding agent is part of the package. That is a big advantage. Claude Code feels more focused, but Codex feels more complete. And for a lot of people, complete is more useful than focused.
Now, let's talk about Kimmy. Kimmy is the one I would describe as strong, but more specific. The Kimmy plan gives you access to Kimmy K2.6, and Kimmy good at generating code. It can be good at solving specific problems, but the consistency is not always on the same level. And when I'm paying for a coding plan, consistency matters more than occasional brilliance. Because in real coding work, you do not just need the model to be impressive once. You need it to be reliable across boring tasks. You need it to edit the right files. You need it to follow instructions. You need it to not overcomplicate simple fixes. You need it to write tests when tests matter. You need it to understand when not to touch unrelated code. You need it to be stable for front end, back end, debugging, refactoring, and product work. That is where Kimmy is good. But I do not think it is the safest first pick. I would still keep an eye on it because Kimmy models have been moving fast. But in this specific comparison, I would put it behind Codex and GLM 5.1 for everyday coding.
So, if I had to simplify the whole comparison, I would say it like this. Claude Code is still the premium classic option. It has a great workflow. It has strong models. It has a lot of mind share. But the price and limits make it harder to recommend broadly. Kimmy is the interesting challenger. It is capable and worth testing. It is capable and worth testing, but I don't think it is the best main option unless you specifically like the Kimmy model behavior or the pricing works perfectly for your use case. GLM coding plan is the flexible option. It gives you GLM 5.1, which is genuinely strong, and it lets you use that model across multiple coding tools. That flexibility is the killer feature. Codex is the most complete option. It gives you strong coding ability, a growing coding workflow, ChatGPT integration, cloud work, review features, and a broader AI ecosystem around it.
That is why this is not just a model comparison. If we were only talking about raw front end output, Claude might still win in a lot of cases. If we were only talking about using one model inside many coding tools, GLM would be the obvious pick. If we were only talking about the biggest general AI bundle with serious coding included, Codex would be the obvious pick. If we were only talking about a strong alternative model to experiment with, Kimmy would be very interesting. But most people are not buying four subscriptions. Most people want to know where to put their money. And for that, I would narrow the answer down to two. CodeX and GLM-5.1.
CodeX is the one I would choose if you want the full AI ecosystem around your coding workflow. It is good for people who already use ChatGPT, want strong code base understanding, want cloud tasks, want review features, and want one subscription that does more than just coding. GLM-5.1 is the one I would choose if you want flexibility and strong model performance inside the coding tools you already like. It is good for people who use Kilocode, Claude Code style workflows, Cursor, Klein, or other agent tools, and want a model they can carry across those workflows. So, for me, Claude Code is still good, and Kimi is still worth watching. But if I had to recommend the two best options right now, CodeX and GLM-5.1 are the best options.
Overall, it's pretty cool. Anyway, let me know your thoughts in the comments. If you like this [music] video, consider donating through the Super Thanks option, or becoming a member by clicking the join button. Also, give this video [music] a thumbs up and subscribe to my channel. I'll see you in the next one. Until then, bye. >> [bell]