People should realize by now that they aren't giving out thousands of dollars worth of compute in a $20 subscription out of the goodness of their hearts. The base level plans aren't meant for any kind of serious work beyond the equivalent of Google searches. At best treat it like a trial.
Session limits are obnoxious and make Codex much less useful for me. I tend to code in spurts when I find some time and the $20 weekly limit was reasonable for my side projects.
I'm now hitting the session limits which means I can either purchase another plan, upgrade, or move off of Codex.
The equivalent alternative is not "no hourly limits", it's simply less overall weekly allowance for Plus users, or higher prices overall. It's a way to balance a scarce resource, and maybe the least frustrating one?
People have a fair chance to learn resource management in 5 hour chunks, while not being limited to silly models, and burning through all tokens in the first hour of the week. Seems mostly good.
Once they capature enough marketshare prices and restrictions are both going to skyrocket. The difference between the subscription usage and API pricing are stark.
This would need to be a collusion across most of the providers (which I think is likely to happen) otherwise OpenAI would get ditched for Anthropic or vice versa, depending on who raises the prices more. If it happens across the board then enterprises that can use Chinese models won’t have that many options but to pay up.
This is interesting to think about. There's so much competition among the frontier labs and from outside via open source it just doesn't seem obvious to me they could maintain elevated prices
Can you talk more about what changed in the last two months? I use open router for when I run out of limits, and I’ve done better than buying an additional sub for my uses. I thought about looking at a more niche player, but it’s not a priority for me yet.
First of all, it still has the 5 hour limits, which is what this thread is about and what I think what most in this thread want to avoid.
Second, the offerings are subject to change at random. They advertised $60 of usage for $10/month (knowing most users wouldn't reach that). This was consistently the case up until early August, when the per-model "usage multipliers" started taking over. Some models give $15 of usage per month, others $30, still others remain at $60, and apparently one at $100 now?[0] Either way, I don't want to expend the mental effort to track which model is the best deal for capability and usage.
Right now I use Hyper[1] and give $100/month to OpenAI. Not OpenAI's biggest fan, but the value is good right now, and that's what matters.
This is one of the reasons I’ve built a resident daemon on top a local CC/codex. 5h sessions is a [lazy] way to shift the scaling responsibility onto a user, so they are staying with us for a while
Clearly, the prices and limitations will increase with time if they don't find an efficient way to perform training and inference of LLMs. It would be great to see those issues being seriously addressed and, eventually, being fixed for good. THAT would definitely make an important and practical difference. Not only economically but also scientifically.
In all of them I've seen it's a window that starts when you haven't been using it in a while and begin work OR when your previous window expired. The window starts with a set number of tokens and cuts you off when they're used, it is fully reset at the end of the window (it's not a sliding window).
When you haven't used any tokens for an extended amount of time it reverts to the point where whenever you send your first token is when the window begins.
I dislike this approach. In my opinion it is much useful a 1x/2x billing based on hour like Deepseek does. Being unable to use it at the hours I need it makes me want to remove the service, not upgrading it.
I wouldn't be surprised if every $10 you add to the subscription cost halves your audience.
I'm now hitting the session limits which means I can either purchase another plan, upgrade, or move off of Codex.
People have a fair chance to learn resource management in 5 hour chunks, while not being limited to silly models, and burning through all tokens in the first hour of the week. Seems mostly good.
What's not good for the customer is the constant change of rules. It is bait & switch.
Second, the offerings are subject to change at random. They advertised $60 of usage for $10/month (knowing most users wouldn't reach that). This was consistently the case up until early August, when the per-model "usage multipliers" started taking over. Some models give $15 of usage per month, others $30, still others remain at $60, and apparently one at $100 now?[0] Either way, I don't want to expend the mental effort to track which model is the best deal for capability and usage.
Right now I use Hyper[1] and give $100/month to OpenAI. Not OpenAI's biggest fan, but the value is good right now, and that's what matters.
[0] https://opencode.ai/docs/go/
[1] https://hyper.charm.land
When you haven't used any tokens for an extended amount of time it reverts to the point where whenever you send your first token is when the window begins.
I've had few weekends where I spend the full week credits over night then have nothing to do for a week