I would not look only at the model, but at the workflow around it.
Claude is still very strong for reasoning over larger code changes, but the cost becomes painful if you use it as a constant pair programmer. For daily work I usually get better results by combining a cheaper model/tool for small edits with a stronger model only for architecture, refactors or debugging.
The real missing piece for me is not another chat UI, but a clean way to make repos more predictable: detect the stack, know how to run tests, know how to build, and avoid guessing commands. Most AI coding tools still waste time because they do not understand the project’s operational surface before editing.
So my answer would be: use Claude selectively, but invest more in tooling around the repo itself.
I've looked into OpenCode and didn't realize that pretty much all of the models are quantized. This makes sense when you consider they offer considerable value compared to competitors, but this seems to reinforce the adage that "you get what you pay for".
I would not look only at the model, but at the workflow around it.
Claude is still very strong for reasoning over larger code changes, but the cost becomes painful if you use it as a constant pair programmer. For daily work I usually get better results by combining a cheaper model/tool for small edits with a stronger model only for architecture, refactors or debugging.
The real missing piece for me is not another chat UI, but a clean way to make repos more predictable: detect the stack, know how to run tests, know how to build, and avoid guessing commands. Most AI coding tools still waste time because they do not understand the project’s operational surface before editing.
So my answer would be: use Claude selectively, but invest more in tooling around the repo itself.
Haven't tried, but have you checked OpenCode?
I've looked into OpenCode and didn't realize that pretty much all of the models are quantized. This makes sense when you consider they offer considerable value compared to competitors, but this seems to reinforce the adage that "you get what you pay for".
Second this. They offer incredible value for the price.
Codex, btw, seems to offer far more limits than claude for the same prices. Their 5.5 model, in my opinion, is on par with opus 4.7.
Have you tried asking Claude?