AI race not China vs US, but closed vs open weight

[ergophobe]

https://fortune.com/2026/08/04/has-the-ai-race-shifted-from-u-s-vs-china-to-open-vs-closed/

I keep seeing that in many regions (e.g. Africa), AI usage is almost entirely cheap open weight models.

[rcjordan]

>almost entirely cheap open weight

Feeds say bet on open weight for the win.

[Travoli]

I recently heard this described as another Apple vs Android... OpenAI and Anthropic are the premium Apple equivalent. Open weight/source models are the Android equivalent.

[littleman]

I am hoping for more of a MS/Apple vs. Linux situation.

[ergophobe]

I think that's more where we are LM.

Kimi is close to Claude and 1/3 the price and other OW models (DeepSeek) are both less good and less expensive. Depending on the task, a model that matches Claude from a year ago might be enough.

https://www.together.ai/blog/kimi-k3-vs-claude-fable-5-on-deepswe-cost-and-coding
https://www.unsupervised.com/resources/what-agent-work-costs

They aren't "free" in either of the senses that Linux is free, but if you are talking about Linux from a TCO perspective, then it isn't free in the beer sense either.

[ergophobe]

See the graph in this article

https://thenewstack.io/open-weight-models-frontier-costs/

[ergophobe]

https://jamesoclaire.com/2026/06/25/the-unbearable-cheapness-of-open-weight-models/

[ergophobe]

"Silicon Valley startups are setting up open models, with some operating on shoestring budgets because of limited interest from VCs"

https://www.wsj.com/tech/ai/the-race-to-build-an-american-alternative-to-cheap-ai-from-china-2e99a28a