> $150/Mtok input, $600/Mtok output
What use case could possibly justify this price?
What use case could possibly justify this price?
Now to have that delivered to you in less than an hour? That’s a huge win.
IR35 rules largely closed down a lot of the high-value contracting business and pushed day rates down heavily.
So why are you saying we'd get a dev for a week for $600? :)
Even we take that at face value, now we're claiming devs are $30K/year. (This was sort of baseline full-time pay when I was in high school, 20 years ago)
Why do I need 1,000,000 tokens for 500 loc?
I don't think further futzing with the numbers makes this work, it's off by multiple OOMs while barely credible
As we all know it's never just 500 lines of code, and 500 lines can take up quite a lot of tokens, especially on a reasoning model like o1-pro which we know can spend quite a hefty chunk of tokens on thinking.
Once you add in iteration and testing you can find yourself racking up quite the bill even on small projects.
It's non-trivial work that requires domain knowledge, i.e. it would have been a 4-6 week project for a noogler, and I would have been impressed if they did it without heavy guidance. (Emergent requirements around purposely using an older API, stuff like that).
- 80 chars per line, 30 occupied (avg'd across 300 KLOC in codebase)
- 500 lines of code
- 15000 characters
- 4 chars / token
- 3750 tokens output
- 10 full iterations, and don't apply cached token pricing that's 90% off
- 37,500 tokens req'd in output
- $600 / 1M tokens
- $0.60 / 1K tokens
- $18
Further we also aren't counting input here which can get long since it includes the previous output, which for the last request will be 33,750 + reasoning + any prompts, which will increase your cost quite a bit there.
But yes that is more reasonable than I'd expect I must admit, but I still think it needs to be at least an order of magnitude cheaper to compete against the other models out there.
I'm not sure I know a lot of employees who would allocate that sort of constant funding to a consumables tool for an employee, given that's your usual monthly cost for a typical saas product.
You generally can't see this due to all the middle men and bloat, and that no corp really wants to measure dev productivity against revenue like this as it'd raise a lot of uncomfortable questions by both devs and shareholders alike.
YMMV
Go check out flutter_pcm_sound_fork, find me even one package with the same streaming PCM => speakers functionality, and I'll give you $500. All I ask is, as a personal favor to me, you read the part in the Hacker News FAQ about "coming with curiosity"
What you cannot do. You may not use our Services for any illegal, harmful, or abusive activity. For example, you may not: Use Output to develop models that compete with OpenAI.
Presumably you run the risk of getting banned if they realize what you're doing.
OpenAI trained on the world's data. Data they didn't license.
Anyone should be able to "rip them off" and copy their capabilities on the cheap.
This reads as if they consider developing models that compete with OpenAI as illegal, harmful or abusive. Which is crazy. (The other dot points in their list in the linked terms seem better).
1) Why wasn't OpenAI doing it themselves?
2) This means we've reached technological singularity if AI models can improve themselves (as in getting a smarter model, not just compressing existing ones like Deepseek)
As far as I know, OpenAI has been doing this, using both experts and Kenyan workers as well as their own discriminator models. Unfiltered synthetic data is generally used more for distilled models and fine tunes for a specific use case.