One big advantage I’ve found — people get attached to models (including me). With open models if you find one that works perfectly for you but the next version doesn’t, you can run the old one forever (or someone will for you)
But… the models will fall behind. As libraries and languages and tool calling updates or the world knowledge changes, the models decay.
Personally, I don’t like the change, but it’s just how technology works so I’d rather move with the flow than try to stick my foot down and freeze time.
Yes but why does that matter?
If I am happy with its capabilities now, I will continue being happy with its capabilities in the future.
Yes, it cannot do the newest magic shit, but why does that matter?
It can still do everything that existed up until that point, which is _a lot_.
Eventually, you might also need something new, but it's not like the world shifts over all problems that exist from <old> to <new> and any tech for <old> problems suddenly becomes obsolete?
Why pay a monthly fee when you can pay for exactly the # of tokens you actually consume?
The API rates are very affordable once you start to optimize for the fact that prepaid tokens seem to massively outperform other kinds of tokens.
I can often do with 1 million tokens what my peers have failed to do with 100 million. For me to spend more than $200/m in prepaid API tokens I'd have to pull a 007 work schedule.
One reason might be request limits. OpenAI's ChatGPT Plus w/Codex ($20/month) provides a worst-case 5-hour-request-limit of 15 for GPT-5.5, 20 for GPT-5.4, 60 for GPT-5.4-Mini. Whereas Z.ai Lite ($18/month) provides a worst-case of ~80 for GLM 5.2 (off-peak; on-peak is 2am-6am New York time). So Z.ai can provide higher limits for a cheaper price. (https://codeberg.org/mutablecc/calculate-ai-cost/src/branch/...)
I don't see any indicator of that happening. And actually token count pricing is frequently being replaced with "credits pricing", and subscriptions with obscure variable limits
the pricing page doesn't seem to call it out anymore, but the claim on z.ai coding plan used to be 3x the usage of the equivalent-price claude plan. whether that's accurate i don't know, but just based on api pricing GLM is way cheaper.
After all the articles calculating OpenAI and Anthropic giving heavily subsidizing their subscriptions, how does OpenCode Go manage to be even cheaper?
OpenAI and Anthropic are trying to pay off a half trillion dollars of investment. They also have the most demand right now, to the point that Anthropic sometimes doesn't have enough compute and that means more limits. They can't stop taking new customers, though, because the market would hate it.
An open weight inference provider only needs to pay for GPUs, or discounted APIs from 3rd party vendors. Same basic financial model but they didn't spend a trillion dollars so their loss isn't as high so they can afford to do more inference for less money, and their demand isn't as high so there's more than enough compute.
One big advantage I’ve found — people get attached to models (including me). With open models if you find one that works perfectly for you but the next version doesn’t, you can run the old one forever (or someone will for you)
But… the models will fall behind. As libraries and languages and tool calling updates or the world knowledge changes, the models decay.
Personally, I don’t like the change, but it’s just how technology works so I’d rather move with the flow than try to stick my foot down and freeze time.
> But… the models will fall behind.
Yes but why does that matter? If I am happy with its capabilities now, I will continue being happy with its capabilities in the future.
Yes, it cannot do the newest magic shit, but why does that matter? It can still do everything that existed up until that point, which is _a lot_.
Eventually, you might also need something new, but it's not like the world shifts over all problems that exist from <old> to <new> and any tech for <old> problems suddenly becomes obsolete?
4 replies →
No problem, "AI" will just write its own frameworks and libs then!
This is a good point I never thought of. I appreciate it.
Why pay a monthly fee when you can pay for exactly the # of tokens you actually consume?
The API rates are very affordable once you start to optimize for the fact that prepaid tokens seem to massively outperform other kinds of tokens.
I can often do with 1 million tokens what my peers have failed to do with 100 million. For me to spend more than $200/m in prepaid API tokens I'd have to pull a 007 work schedule.
> Why pay a monthly fee when you can pay for exactly the # of tokens you actually consume?
Because my 500m tokens so far this month would cost me about $500. My subscription is 100$/month.
That’s insane. 500m tokens costs me $12 on Deepseek.
One reason might be request limits. OpenAI's ChatGPT Plus w/Codex ($20/month) provides a worst-case 5-hour-request-limit of 15 for GPT-5.5, 20 for GPT-5.4, 60 for GPT-5.4-Mini. Whereas Z.ai Lite ($18/month) provides a worst-case of ~80 for GLM 5.2 (off-peak; on-peak is 2am-6am New York time). So Z.ai can provide higher limits for a cheaper price. (https://codeberg.org/mutablecc/calculate-ai-cost/src/branch/...)
Subscriptions are done. By the end of 2026 everyone will be paying for actual mils of tokens consumed, via API calls.
I don't see any indicator of that happening. And actually token count pricing is frequently being replaced with "credits pricing", and subscriptions with obscure variable limits
https://news.ycombinator.com/item?id=48618455
I pay month to month.
the pricing page doesn't seem to call it out anymore, but the claim on z.ai coding plan used to be 3x the usage of the equivalent-price claude plan. whether that's accurate i don't know, but just based on api pricing GLM is way cheaper.
OpenCode Go is $10/month and the limits are much more generous than those or Codex
After all the articles calculating OpenAI and Anthropic giving heavily subsidizing their subscriptions, how does OpenCode Go manage to be even cheaper?
OpenAI and Anthropic are trying to pay off a half trillion dollars of investment. They also have the most demand right now, to the point that Anthropic sometimes doesn't have enough compute and that means more limits. They can't stop taking new customers, though, because the market would hate it.
An open weight inference provider only needs to pay for GPUs, or discounted APIs from 3rd party vendors. Same basic financial model but they didn't spend a trillion dollars so their loss isn't as high so they can afford to do more inference for less money, and their demand isn't as high so there's more than enough compute.
Because it offers cheap open source models, not GPT and Claude. I mentioned it as an alternative to Z.ai's subscription in OP's comment, not to Codex.