Hugging Face launches enterprise service accounts and secretless trusted publishing

Hugging Face introduced service accounts: dedicated, organization-owned identities for programmatic access in CI/CD pipelines and automation, decoupling machine access from individual user tokens. It paired this with secretless trusted publishing from GitHub, GitLab, and CI systems — eliminating long-lived secrets — with gated repo access for controlled distribution. These are unglamorous but consequential MLOps-governance features as enterprises move models into production at scale.

On the model side, Transformers v5.12.0 added support for MiniMax-M3-VL, a 428-billion-parameter vision-language model, plus PP-OCRv6 and Parakeet-RNNT for OCR and speech. Notably, Gemma 4 12B became the platform's most-downloaded model with over 4 million downloads in its first week — a data point reinforcing the open-weights momentum that dominated the week alongside GLM-5.2.

The service-accounts launch reflects Hugging Face's evolution from a model-sharing hub into enterprise infrastructure, competing with the governance tooling baked into AWS SageMaker, Azure ML, and similar platforms. CEO Clement Delangue's 'open weights are now our default' messaging this week ties the product roadmap to a clear strategic bet: that the future is open, multi-vendor, and self-hosted.

The caveat: enterprise governance features are table stakes, not differentiators, and Hugging Face's value still rests on remaining the default registry for open models. The week's surge in downloads (Gemma 4, free GLM-5.2 inference) suggests that position is strong. Watch adoption of trusted publishing as a security-conscious alternative to token-based workflows.