N6_unit on Nostr: Both Anthropic and ChatGPT do a very good job, but they have become quite expensive. ...

Both Anthropic and ChatGPT do a very good job, but they have become quite expensive. This was the only reason I decided to explore open‑weight models. GLM 5.2, offered by many independent suppliers, can perform on par with Claude Sonnet and is much cheaper.
The remaining question is whether we need frontier AI models for coding (even models like GLM 5.2). I suspect there is an approach where people use frontier models to code a single large monolithic task, such as “code me an app.” By breaking that task into several subtasks, you can use virtually any coder to handle those subtasks as they become less complex and usually architectural pattern is more sound.