Hugging Face runs the central platform where the machine-learning community shares, runs, and builds on open models, datasets, and demos, and maintains the Transformers library that became the standard way to use AI models.
Notable milestones from Hugging Face’s journey — funding rounds, launches, and other moments worth knowing.
LeRobot v0.6 added world-model policies, a reward-model API, six unified simulation benchmarks, and a rollout CLI that records human corrections as new training data. The release turns Hugging Face's robotics stack from a model-and-dataset library into a tighter loop for training, evaluating, deploying, and improving robot policies.
SourceHugging Face and Cerebras paired Gemma 4 inference on Cerebras with Parakeet speech recognition and Qwen3TTS to demonstrate an inspectable, modular speech-to-speech pipeline. Hugging Face says the same pipeline already powers Reachy Mini robots, with more than 9,000 robots in the wild.
SourceQualcomm expanded its Hugging Face partnership to map Hugging Face storage, inference and developer workloads onto Dragonfly data-center systems, while onboarding the Hub's 3M+ open models across phones, PCs, vehicles, industrial devices and racks. The companies also plan a Hugging Face Agent for hybrid orchestration between local Qualcomm-powered devices and cloud infrastructure.
SourceHugging Face launched an app store for Reachy Mini with around 200 downloadable or modifiable apps, aiming to make open-source robotics approachable beyond researchers. Axios reported roughly 10,000 devices were already in customers' hands or on the way.
SourceHugging Face acquired French startup Pollen Robotics and its Reachy humanoid robots, betting that robotics needs the same open, shared foundation as models — and reconnecting the company with its French roots.
Hugging Face crossed a million hosted models, on its way to millions more alongside datasets and runnable Spaces. The Hub had become the default place open AI is published, found, and run.
In what Delangue called its largest acquisition to date, Hugging Face bought Seattle-based XetHub — ex-Apple ML infrastructure engineers — to replace Git LFS with chunk-level storage built for the terabytes of models and datasets flowing through the Hub.
SourceSalesforce led a round joined by Google, Amazon, Nvidia, AMD, Intel, Qualcomm, IBM and Sound Ventures — rivals who all depend on the Hub lining up behind it. The raise doubled Hugging Face's valuation and made it a clear AI unicorn.
Salesforce VenturesThe output of BigScience — a year-long collaboration of over 1,000 researchers Hugging Face convened — BLOOM matched GPT-3's scale across 59 languages and was released openly, a deliberate counter to the closed frontier labs.
SourceLee Fixel's Addition led the round as the Hub was turning into the place the ML community shares models and datasets. The same year, Hugging Face folded in Gradio to let anyone wrap a model in a shareable web demo.
By open-sourcing the tooling behind models like BERT and GPT-2, Transformers turned cutting-edge research into a few lines of Python. It quickly became the default way researchers and engineers load and run models, and the foundation of everything Hugging Face built next.
SourceClément Delangue, Julien Chaumond and Thomas Wolf founded Hugging Face in New York — named after the 🤗 emoji — as a playful teen chatbot. They soon realised the open-source ML tooling underneath was the real product and pivoted to it.
Source
