I'll be amazed if they manage to keep their infra responsive over the next 2 weeks.

I've been getting a lot of these messages today:

API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited

They just leased a massive spacex data centre.

Even so. The 2 week period will predictably unleash a feeding frenzy.

Limited "free" time is what game developers do if they want to stress test the infrastructure code until it breaks.

Yeah that's how I'm using it right now. Smoke em while you got em...

No issues that I've seen so far. Seems to be holding up for now.

Opus will be gutted furthermore. /s I feel 4.8 is very slow in last 2 days

This is the entire business model of all AI companies. It costs far more to run the datacenters and build more capacity than they could ever hope to make back at current pricing models. I'm looking forward to pricing to catch up with reality and the resulting chaos that ensues.

Kind of how DeepSeek v4 dropped their pricing? I sense a shift which will hopefully bring lower and lower cost. Then again Qwen3.6 coding has been all I've needed for my projects and I'm perfectly fine with free.

Are you paying attention? These companies are trying to get market share without being anywhere close to making a profit - they are heavily subsidized. Many hundreds of billions have already been spent and will continue to be spent until the stupid fucking investors realize they will never get their money back. I have no doubt that day is coming.

The people actually working in this space will tell you that the cost of all of this continues to crater. Anthropic is already profitable minus its intentional forward-looking investments into R&D.

Only if you count the 2 months of discounted compute Musk gave them.

I'm amazed people still believe this narrative after all the clear evidence to the contrary

You're the one whose wrong here and you will not reap what you didn't sow.

That's fine but can you please post some facts, data or a view point rather than just you're wrong. We could value from your ideas.

"already profitable minus", so, not profitable?

Serious investors look 10 to 20 years in the future. Everyone used google and youtube in 2006, but youtube wasn't profitable til 2016. How could a business burning money by hosting video ever be profitable? Costs come down BUT THEY JUST ADDED 720p, cost comes down, BUT THEY ADDED 1080p, cost comes down, 4k! cost comes down.

IMO the data from chats alone is worth $200B to Google.

but they are trying to IPO with 2-trillion dollar valuations

The amount invested into AI companies is no where near anything we've ever seen before. It's apples to oranges.

Not really a comparison when the spend on YouTube was x10 smaller, and Googles core business has always been profitable beyond any hobby spending on YouTube.

The investors will get their money back on the IPO. They'll dump all their stocks in the market and run away, leaving retail with the bill.

I was just thinking this reminds me of the scene in The Wire where Avon admits to D'Angelo that the new heroin is in fact just the old heroin with different baby powder cutting it.

I was just saying last week: If Opus 4.8 max is as good as we get, and we plateau there, I think I'd be fine with it.

For the stuff I've thrown at it, that configuration has done a really great job. Including 70+KLOC go proxy with extensive test suite, some retro games, and more.

Seems to me this is more honest than the Mythos claims a while ago. too powerful to release publicly. Too expensive?

Didn't they admit this at the time? Cost was one of the reasons they gave for not immediately making it public.

Or maybe its all about compute availability like they say. It could be that they plan to start training a new model on the 22ed, so the amount of compute available for inference will be greatly reduced.