Hacker News Reader

Comment by peri-cl | Hacker News Reader

> "Makes me wonder what the next headline is going to be, probably some deal between NVidia and the federal government over export restrictions or something."

Chinese models disappearing from HuggingFace (recently acquired by Nvidia).

GLM-flash and Qwen-next disappeared from ArtificialAnalysis (the top benchmark site) a few days ago; I was wondering if that was a prelude to something greater. Testing the waters.

View on HN

That's likely new coming PRC sanctions on exporting LLMs, not something Nvidia did. https://sanctionsnews.bakermckenzie.com/china-proposes-expan...

Gee that's pretty much an antitrust case right there. Sell NVidia while you can.

Um, lol, no.

For one, the US doesn't do anti-trust these days, just pay off uncle-trump to get your way.

Two, there are tons of model providers.

"These days" are ending. Nothing about the current geopolitical climate, global supply chain, energy price, or socioeconomic reality of inter-generational quality of life chasm is stable. Taking a trillion dollar bite out of a bubble beneficiary is a clear opening move for a novel government with restored oversight and an eye on the books.

The models are still on AA, not sure what you mean

They're not? They are no longer benchmarked, and not available on any of the charts (not even manually).

Their benchmarks have been replaced with "N/A" for several days now,

https://artificialanalysis.ai/models/qwen3-8-flash-next

https://artificialanalysis.ai/models/glm-5-3-flash/

Oh I see, sorry, I just went as far as checking they were still on the site, didn't check the details.

That is really weird given how widely used they are.

AA updated their benchmark a few days ago and haven't retested every model yet, you can scroll down the page and see those in the old benchmarks

I'm pretty sure that's not it. They had 30+ models benchmarked and ready at launch, for a seamless transition. We're talking about the charts that're the hero images on their landing page.

These comparison charts missing two of the most popular models are defective.

(Also, these two happen to be "flash" models, which are among the fastest and cheapest to run benchmarks on. Pennies per task).

Reply on news.ycombinator.com