HF co-founder: Most-derived model on the Hub isn't Llama—it's Germany's FLUX.1-dev
robrombach · x · 2026-10-20
Hugging Face co-founder Thomas Wolf shared data responding to Mistral CEO Arthur Mensch's "Build or surrender" essay, which argues Europe must not cede frontier AI to the US and China.
Wolf's counterintuitive fact: the most built-upon model on the HF Hub is not Llama or Qwen but FLUX.1-dev from German lab Black Forest Labs, with 43k derivative models.
He adds that Europe trained 8% of open-model compute on the Hub, yet 18% of what people build on top of open models is European. His takeaway: build, and build open. Both will appear at GTC Berlin this week.
More from Companies & People
- PyTorchCon enterprise day features executive sessions with Red Hat, NVIDIA and IBM — PyTorch · 2026-10-20
- a16z: mobile era says chips led first, but AI apps and services may win in the end — a16z · 2026-10-20
- Mystery model Jev hits 1 trillion tokens a day, a quarter of the Fortune 500 in five weeks — wandb · 2026-10-20
- Open Source AI Awards Night Lands in SF Oct 23 with Unsloth, Hugging Face Speakers — latkins · 2026-10-20
- AI and the vanishing first rung: firms quietly stop backfilling junior roles — LexSokolin · 2026-10-20
- Berkeley AI Risk series hosts talk on where AI safety goes after 'Coxon Day' — wfithian · 2026-10-20