Tell HN: OpenAI $500 ProMax plan listed in API

3 days ago

OpenAI's API is showing the first signs in the API of a new 500$ ProMax plan

https://chatgpt.com/backend-anon/checkout_pricing_config/con...

I wonder what's the ceiling for these plans' prices. At some point, it becomes a non-negligible fraction of the salary for the employee, especially in non-US employment markets.

  • A lot depends on how good open source models do. I know I can buy a 20k rig or whatever to get model X. If someone wants to charge me 3k a month to write code I just buy the hardware.

    • I’m interested: how would you spend $20k to get close to performance parity (speed and ability) with a frontier model?

  • I think there will be an employer backlash at some point. Paying an extra couple thousand in subscriptions so a junior can write copy-paste emails and Slack messages? Nah fam, write those yourself.

    • Or just use AI mode on Google's homepage. I asked it to find an article that compares a certain thing to another, which doesn't exist. So Google automatically wrote a 5 page article comparing the two things and built a proper article with cited references. Most people don't need a subscription to anything.

      1 reply →

    • In my observation employers are willing to throw practically unlimited cash, probably because it’s directly funding a future where they don’t have to pay pesky engineers with their holidays and their pensions.

      2 replies →

    • Of course a company doesn’t want to pay for someone to delegate all their work for an additional fee.

      This is why some companies stopped hiring juniors. They would rather pay a mid-level or higher with the expectations that the LLM use will be a productivity booster in a more direct sense. So the bar also rises.

Cloud agents in Codex are quite poor solution compared to Claude Code, you are forced to use your own computer, which is not the desired setup for anyone with a standard laptop not suitable for doing lots of work in parallel.

So I would still prefer to spend some dollars in Anthropic or even Cursor solutions, which work much better as cloud dev envs.

If it's unlimited that would be tempting. But that will never happen and if it does "unlimited" would not be

  • Particularly in this situation, it cannot be unlimited. A command to one agent could scale into a whole datacenter very quickly.

    It can feel unlimited for many use cases, probably. But I doubt it can, let's say, 3D model things in blender on 3 different computers for 24/7. And hey, why not 30 things? 300? If you're just watching them, it's easy to get ridiculous.

    • I understand your point; but I disagree - Unlimited is possible.

      It gets into a "define unlimited" discussion. You could give "unlimited" usage, but still "limit" both concurrency and throttle speed. Even if you run 24/7, they still have knobs they can twist.

      3 replies →

to be honest wasn't much surprised cause of the current situation like frontier models being so expensive and the models like opus consuming way much more tokens you just can't justify any of that. this token math is just so damned right now :(

the bottom line is that a.i is getting more and more expensive and they have excuses for that. just last week trae.ai reduced its plan credits from 400$ to 100$

  • Inference for the same level of intelligence is getting cheaper. Frontier intelligence keeps finding ways to get more expensive, yes.

    But gpt-6-sol, for instance, dropped price for the same level of intelligence. That will keep happening, and someday we'll have basically-free sol. And hopefully by then, very cheap astra.

    But some new wild thing will be very expensive.

    I'm sure I'll figure out how to use the new one, but I would be extremely happy with near-free Astra.

  • Claude subs are great value again ever since Astra release moved lots of people to OpenAI subs. Even moreso since Opus 5.5 release.