We can tell that the inferencing costs for many of these models are low enough that these models are being sold close to real costs on the basis that many of them are open weight and available from third party providers who have no incentive to subsidize them.
I think the frontier labs will need to drop their high per-token prices at least for their low and mid-level models for the reason that several Chinese models (at least Qwen, DeepSeek, Kimi and GLM) are "close enough" that with the right harness they are cost effective alternatives.
They won't necessarily need to close the gap - at least not yet -, because these models won't necessarily compete at the same token counts. E.g. at least some of them need to do far more work to solve the same problems.
But, yeah, the prices will come down one way or the other.
At the same time, even the subscriptions for the cheap Chinese models are probably subsidised, and those subscriptions are likely to get less generous over time.
I really doubt Deepseek is subsidised. It's roughly the same price everywhere you look. Deepseek is using the Huawei hardware (as far as I managed to understand from various articles) and hence the savings.
Don't know why people keep parroting this, this is incorrect. Chinese electricity prices are equal or slightly cheaper then most of North America. But significant pockets such as those around the Quebec or other hydro plants are significantly cheaper then Chinese power pricing.
Not only that, China may subsidize AI, but so does the US.
Quebec has lower rates then 7¢/kWh at data center / wholesale level. Quebec spot market runs negative sometimes, apparently. And Oklahoma has cheap power, and probably other places. Not sure your utility bill is the place to get accurate numbers.
"Mean wholesale electricity prices in 2024 were lowest in SPP ($27.87/MWh), the Southeast ($29.72/MWh), and Southern California ($29.95/MWh), and highest in the Northwest ($59.98/MWh)."
Okay interesting. I presume that China also has low cost areas too no? Their grid at least seems more stable. Datacenter construction is more likely to raise prices in the US than there.
China's grid has had some serious issues over the past decade that didn't get widely reported for all the reasons you can think of. Some of them were exasperated by poor planning and censorship making it hard to hold anybody accountable. Not to say that they don't/didn't eventually work on it, but there was a widely held belief that the people at the top weren't even aware of the issue until foreign firms were directly impacted. This is not to say they can't or won't expand come hell or high water, though.
Yeah, this argument is bullshit. You can head over to Openrouter and look at the token cost for deepseek-v4-flash and deepseek-v4-pro. They are very competitive on the open market
I think the frontier labs will need to drop their high per-token prices at least for their low and mid-level models for the reason that several Chinese models (at least Qwen, DeepSeek, Kimi and GLM) are "close enough" that with the right harness they are cost effective alternatives.
They won't necessarily need to close the gap - at least not yet -, because these models won't necessarily compete at the same token counts. E.g. at least some of them need to do far more work to solve the same problems.
But, yeah, the prices will come down one way or the other.
At the same time, even the subscriptions for the cheap Chinese models are probably subsidised, and those subscriptions are likely to get less generous over time.