Commenting actual Models and coding-Plans

[Theo Gottwald]

In the following i will give you some commetns on coding plans by 21.06.2026

1. Open AI Coding Plan:
Expensive, PRO-Plan 100$. But still the best for all types of work, if its difficult.
This is still the standard for daily work. even browser usage is now very good.
Lets me give you an example. I told it the problems with IONOS.
It logged into the IONOS Page and into the Forum. It repaired everything autonomously.
Normally i would have needed to hire someone for that. It will just do it.
Let me add that the support from OpenAI is zero.
Whatever problems you have - be sure you get no help othe then a bot-chat-window and sometimes even that does not answer. (With M3 you can even write a Mail to the owner and get an answer).

2. GLM Coding Plan from Zhipu-AI
The GLM-Model is from all Open Source Models the only model that cann 100% compete with
OpenAI in difficult tasks. But its very strange expensive. I tried the lite and the pro Plan.
The API is melting like a wax-candle after some minutes nothing is left.
The numbers of used Context Bytes skyrocket in a way that can never jappen with the displayed
currently used context. In my personal opinion this thing is either damaged and doesn't count cached Token properly, or its not recommended by far most expensive. Even much more expensive then Open AI.

3. Minimax Coding Plan.
This is just the opposite. Minimax is not the strongest Model when it needs to use a large context window. For smaller Tasks its really Top. Especially the Coding Plan is from my testings the most efficient. I used it wherever possible, trying to use all available Tokens, gave it Tasks, let it run as Subagent for kimi and for GLM and for OpenAI. They like to use it - but no chance, i will end this week with not all Tokens used. If you want a cheap coding plan then this is for you. You can try the performance of M3 on their web-page and check if its ok for your tasks.
Where i used it? It has made all the concepts and raw code for the new enhancements for PBXB64. The Implementation was done by other models because that a Task involving a large codebase.

AND HERE you get it -12% cheaper!
MinimaxM3 Coding Plan

Kimi K2.7
This is the model doing most of the work here. From its analytical power its sitting between GLM's 5.2 and Minimax M3. So its doing the impmentation of most features. The strong side is its Agentswarm Capability. Especially the yearly plans have a fair price. However to really use Agentswarm you will need the Top-Plan else it will eat up all your API credits.

This time Kimi is special as if you use their Coding App it says "Quota half for this task"
which is not absolutely true. In fact using the App (its not official its just what i see here!)
is currently NOT reducing the Quota at all.
At least here ... Its ... free. Amazing but free.
The App has some disadvantages, you will always need to trigger it, you can not simply make a gola like with the CLI (and put yourself on the sofa ...).
And it uses the Kimi K2.6 currently. But .. its free maybe untill the 30the of this month.

About the weaknesses: Kimi is very optimistic and when you give it something to implement, it will come back and say "Its ready" but some of the used features are just "Stubs" (empty procedures).
You need to handle that. Its just an very optimistic model in any way.



Picture: Using Kimi K2.7 and Agentswarm to have 22 Agents working at once on PBXB64


A bit later ...
Deploying 22 parallel Agents is like using a BIG HAMMER on a coding task. I'd say no other Model can do that.