I subscribed to OpenCode briefly ($5 for the first month) and was ok with it. I say ok, because if you have a full-time job and do this as a hobby, you are unlikely to hit the limits even if you use the expensive models. If this were your full-time job, then you could easily hit the limits.
I am using completely the free tier of everything (this is what I have working thus far):
devin
kiro
copilot
openhands
opencode
codex
cursor
antigravity
jules
aider (openrouter)
cline
kilo
grok
I have my own 'router' which routes to the providers that reset daily and burns through them first, followed by weekly, and monthly. The best performers for me were:
devin, opencode, antigravity, jules
codex, cursor, and claude perform well, but have substantially lower limits (for the free tier).
I have just started using cline and kilo and they seem to be on par with devin, but I don't have much experience.
my experience with grok free was about nearly identical to when I first started with AI in 2024, their free models are terrible. But, I have only used it about 2x, so perhaps I need to try more to prove that wrong?
That said, I feel that the opencode big pickle model outperformed qwen 3 max and qwen 3 plus and I paid for qwen ... Opencode big pickle has generous free limits and I generally don't hit them. But, the thing is, you have absolutely no idea when you will hit the limit.
Eh, the problem now becomes, how do you validate that what AI generated was correct or useful? That is where I spend most of my time these days. I have pipelines, but still, the human is the bottleneck.