Hacker News new | ask | show | jobs
by drnick1 3 days ago
> American labs need to release frontier-grade open-weight models under licenses that startups can actually build on.

To be fair, OpenAI has released a couple of (then very good) OSS models. I run the 20B version at home and it is excellent for reviewing text and common tasks like drafting bash scripts. There is a larger 120B that you can't realistically run on consumer hardware at reasonable tok/s too. I wish OpenAI updated these models more frequently though.
3 comments
stefan_ 3 days ago
"We have been having extensive discussions around open source strategy. [..] one thing we'd like to do soon is to create a language model with the approximate capability of GPT-3 that can run locally [..]. In general, we think this helps discourage others from releasing similarly-powerful models, and makes it harder for new efforts to get funded." - Sam Altman emails OpenAI board, 2.5 years after GPT-3

Chinese release open models to drive the state of the art, OpenAI and crooked Sam do it to keep you down.

link
drnick1 3 days ago
> Chinese release open models to drive the state of the art

Are you sure it isn't just another form of Chinese industrial policy? China does not have frontier labs, but through distillation and their own work they can get pretty close. It's not enough to be competitive with Anthropic and OpenAI, but there is still money to be made by selling compute (software as a service), and in any case it's better than being left behind in the AI race.

link
potwinkle 3 days ago
There are coding tasks where Kimi K3 outperforms Fable (haven't done much comparing between it and Sol) and the Chinese labs have access to their own synthetic datasets, along with frontier research. We're leaving behind the days where Chinese models are distilled Claude, but I hope Anthropic/OpenAI can continue to accelerate.
link
johndough 3 days ago
> There is a larger 120B that you can't realistically run on consumer hardware at reasonable tok/s too.

gpt-oss-120b runs at 30+ tps on Strix Halo and +75 tps on a MacBook Pro M5 Max 128GB.

> I wish OpenAI updated these models more frequently though.

I think the spiritual successor is the Nemotron 3 series, although they also are getting a bit long in the tooth: https://research.nvidia.com/labs/nemotron/Nemotron-3/

The Gemma 4 models are a bit more up-to-date: https://huggingface.co/collections/google/gemma-4

Or Qwen3.6: https://huggingface.co/collections/Qwen/qwen36

link
bigyabai 3 days ago
The OSS models are good, but weren't very competitive at release with alternatives like Qwen3 Coder A3B. If you were a startup, why would you build on OpenAI's first (and last, for quite some time) model versus Qwen3, which was updated with multiple open-weight point releases?

The article certainly has a point, the current release cadence is untenable for startups that want frontier-grade models from American labs. Unless America wants to cede that market entirely, OpenAI et. al. need to kick it into gear, fast.

link