Hacker News Reader

Comment by ingen0s | Hacker News Reader

Wait - you spent north of $400,000?

People love to flex by using the API costs for their subscription tokens.

I personally find it ridiculous. The cost is what you paid, not what someone else could have paid. Markets and all that.

It would be like using spot instances on AWS but flexing by boasting about how much on-demand $$ compute you used.

In this case, I think they are just giving you an idea of what it cost to do this. They aren't saying it's prescriptive

Except.. It doesn't really give you an idea what it cost? That's not what it cost. How much did it actually cost? Who knows!

It's also TheoGG sooooo.. Those influencer instincts tho.

Assuming he used $400,000 in tokens, but used subs

I wonder how much AI companies lost on this?

Edit: not that I’m worried about them, I’m just curious

> I wonder how much AI companies lost on this?

Probably zero, or even a profit. The idea that these companies are making a loss on subscriptions is an unsubstantiated HN fallacy.

On the contrary, they're probably making a small profit on subscriptions, or at least break even, and absolute bank on API pricing. Anthropic recently reported an 80% gross profit margin, for example.

https://www.reuters.com/business/retail-consumer/anthropic-t...

Profitable, if you just ignore how much money they spend

Excited for their IPO just to put this to bed, one way or another.

It might, but also might not. I would imagine the economics are genuinely very tangled up even before you start obfuscating it - it’s very hard to fairly allocate fixed costs between training and inference.

I think a clearer picture would be from inference-only orgs.

What are “subs”?

Subscription

So you pay $20/mo, let it run for a long time, then claim you used the equivalent of $1/2 million?

Pretend you have an unlimited cell phone plan. Talk on it for hundreds of hours a month for free

Then take the minutes you used and compare what it would’ve cost, if you paid by the minute

I think he has 5-10 of $200 subs, but yes, that's the general idea.

Most of it was wasted on OpenAI tokens. It actually cost them $24k USD according to the github description.

So, you could do it for a couple hundred bucks using DeepSeek.

Probably not since Astra was also not smart enough to do it, and Astra is many times smarter than Deepseeks models.

If DeepSeek was smart enough. There's a huge risk that it isn't since the only model that ultimately succeeded was Opus 5.5 (sol and astra were tried before).

No that's the theoretical token value cost based on chaining multiple 20x subs.

$420,000 worth of tokens, though I'm pretty sure he used subs.

Sibling dead comment said it already but that's likely an estimate of what it would cost via API and in practice it was a few hundred or thousand dollars of subscription fees.

> I did over $400,000 in API priced tokens

i'd bet this is a few subscriptions over a few months rather than paying directly for tokens.

This was OpenAI provided unlimited usage for evaluation. The current working version made by Opus was the one that used a couple of 100$ subs

An absurd amount. The cost wouldn't exceed $500 under a subscription model.

I wonder how that cost compares to paying a human programmer to produce the same end result.

Reply on news.ycombinator.com