So your prior is that token costs are only going to go up. Doesn't stuff like GLM 5.2 and Deepseek change this? I.e. something close to Opus 4.5 that runs 10X + more performantly.

▲

esseph 5 hours ago | parent [-]

No, because those loans still have to be repaid.

▲

ianm218 4 hours ago | parent | next [-]

Could you explain your mental model of the situation a bit more?

Let's take something like GML 5.2 that is open source - anyone can run it on NVIDIA or AMD chips. People are currently making money by running GLM 5.2 Nvidia and AMD chips and selling inference on aggregators like openRouter.

So the cost to serve is like cost of Chip or financing for Chip + power and rackspace somewhere. We basically know that each year you will be able to server exponentially more inference per $ on Nvidia and AMD hardware as it gets to newer generations, so why would you expect the cost of inference on open models to also increase?

Like let's say not through a Chinese subsidized provider but like BaseTen https://www.baseten.co/pricing/ why would that get more expensive year over year?

▲

esseph 3 hours ago | parent [-]

> Could you explain your mental model of the situation a bit more?

I'm talking specifically about US-based AI companies / hardware companies and their circular investments and their customer-facing pricing.

> We basically know that each year you will be able to server exponentially more inference per $ on Nvidia and AMD hardware as it gets to newer generations, so why would you expect the cost of inference on open models to also increase?

Because the cost of training continues to increase while getting increasingly harder and harder to make gains. On top of that, there is the delay between investing in the datacenters, staff, running the training, operational costs of customer use, etc and actually recouping profit via customer sales. Your costs go up because their costs go up. Likely even more so now because of the increased risk the US Gov will shut down sales and use of their newest models. That risk will get bundled in with pricing.

▲

ianm218 an hour ago | parent [-]

Hmm but isn't your point moot if the GLM 5.2 type models are good enough to do these large scale port to Rust projects? Like maybe Anthropic goes bust but doesn't really matter for this case.

I feel like your conflating some general skepticism around the trillion dollar valuation of the US majors and their business model with the topic at hand - large scale C to Rust conversions and similar.

	▲	esseph 14 minutes ago \| parent [-]
		[delayed]

▲

croemer 3 hours ago | parent | prev [-]

No, they don't, they can also be defaulted on/written off. Open weights are there to stay, plus research, plus the GPUs, just the power needs to be paid.

There's a limit to price hikes due to competition, and open weights can't collude so closed weight companies don't have that much wiggle room to raise prices unless they are much better than open weights.

	▲	esseph 3 hours ago \| parent [-]
		> There's a limit to price hikes due to competition Unless you say you have the best model in the world, and then everybody will fight to use it. > No, they don't, they can also be defaulted on/written off. Bye, bye world economy.