/Tech1h ago

AI2's Nathan Lambert rejects claims that Western developers must rely on Chinese open-source models, citing rapid US pre-training progress

Story Overview

Nathan Lambert, post-training lead at the Allen Institute for AI, is pushing back on the notion that Western teams have no choice but to fine-tune Chinese open-weight models like DeepSeek or Qwen, instead highlighting ongoing U.S. pre-training gains and a visible pipeline of open-model work stretching through 2026.

526132.6K

#80

Original post

xlr8harder@xlr8harder#1855inTech

I think saying there are no good open source models from the west was clearly true 3-6 months ago, but now there are at least 3 US teams making fast progress and scaling efforts.

We'll see how far they go, but I have high expectations.

Harry Stebbings@HarryStebbings

“We have a crisis of open source models in the Western world.

Outside of China, there are no good open source models. We don't even have any in the US now.

The talent, the capital and the focus to be best-in-class at pre-training, mid-training, post-training is an extremely scarce skillset.

The answer for a lot of countries may just be to take an open weights model from China, post-train or fine-tune their own version, and have that be what they start from." @EverettRandle

Why does the West have such poor open source models and is it a national security threat to rely on open-source Chinese models @demishassabis @ylecun @aidangomez @lilianweng @janleike

9:22 AM · Jun 20, 2026 · 2.2K Views

Open Question

Pipeline keeps multiple U.S. teams in motion

Lambert points to a backlog of planned open-model releases and research milestones that he expects to land before the end of 2026, with at least three American groups now advancing pre-training efforts that were not visible a few months earlier.

Developer Impact

Fine-tuning Chinese models is no longer the default workaround

Recent U.S. releases such as the fully open OLMo 3 family are positioned as competitive alternatives on both base performance and reasoning variants, reducing the immediate pressure for developers to default to overseas open-weight checkpoints.

Sentiment

Users express optimism that US teams will continue creating and funding strong open source AI models.

Pos

100.0%

Neg

0.0%

1 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS564BOOKMARKS2LIKES8RETWEETS1REPLIES1

Nathan Lambert@natolambert

@xlr8harder At this point people need to just read my 2026 backlog.

xlr8harder@xlr8harder

I think saying there are no good open source models from the west was clearly true 3-6 months ago, but now there are at least 3 US teams making fast progress and scaling efforts.

We'll see how far they go, but I have high expectations.

1h56482

xlr8harder@xlr8harder

@xeophon @natolambert I definitely expect they will continue to create and fund strong open source models.

1h421

umumu@umi33563

@xlr8harder I don't think that was true even 3 months ago if you count smaller models (and I very much do). You could say "nothing OSS in the West that at the first glance might be beating SOTA closed chungus LLMs", but this "no good" phrasing is very meh.

29m8