Y Combinator's Garry Tan calls for US open-weight labs to distill frontier models
2026-09-11
Y Combinator's Garry Tan said US-based, smaller open-weight AI labs should use distillation training techniques on American frontier AI labs' models — the same approach Chinese open-weight labs are reported to use — to give the US a more robust set of open-weight options that aren't Chinese-made.
Significance 3: A credible, named-figure strategy signal about the US open-weights competitive position, but a proposal/opinion rather than an action any lab has taken — meaningful strategy commentary at the level the rubric reserves for 3.
Implications · machine-drafted, not owner judgment
Tan is naming distillation off American frontier models as a legitimate strategy for closing the US open-weights gap — a tacit acknowledgment that Chinese labs' rapid open-weights progress (which Anthropic has separately alleged relies partly on distilling its own models) may need an American mirror-image response. If a US frontier lab's models become a distillation target the way Anthropic alleges has happened to it, expect similar IP and terms-of-service disputes to surface domestically.
- A named US open-weight lab publicly adopting a distillation-based training strategy
- A frontier lab changing API terms specifically to restrict distillation following this commentary
Sources
- press Y Combinator's Garry Tan wants US open-weight AI labs to 'distill' frontier models, too retrieved 2026-09-12