Kubeflow pushed toward graduation from the Cloud Native Computing Foundation, using KubeCon + CloudNativeCon Japan in Yokohama to lay out a wave of releases spanning notebooks, pipelines, and distributed training. The open source machine learning platform, built to run on Kubernetes, has formally applied for CNCF Graduation, the foundation's top maturity tier reserved for projects with broad adoption and a sustainable contributor base.
Kale 2.0, the latest release of the Kubeflow Automated pipeLines Engine that turns annotated Jupyter notebooks into production pipelines without hand-written SDK code, is now officially folded into the Kubeflow project rather than treated as an adjacent tool. The release rebuilds Kale around the newer Kubeflow Pipelines v2 (KFPv2) architecture, Kubeflow said.
Kubeflow Notebooks is also nearing a next-generation release, replacing its current setup with a declarative, CRD-driven architecture for provisioning JupyterLab, RStudio, and VS Code environments on Kubernetes; an alpha build is available now. The Kubeflow SDK picked up native Apache Spark Connect support in the same update, letting developers run Spark jobs on Kubernetes without separate infrastructure configuration, alongside built-in blueprints for large language model fine-tuning.
Kubeflow Trainer, the project's distributed-training component, added MPI support to cover high-performance-computing workloads alongside standard AI training jobs. A proposed OptimizationJob custom resource would extend Trainer into hyperparameter tuning, and the community is separately working to add reinforcement-learning algorithms such as GRPO and PPO for large language model post-training.
The project's underlying Community Distribution also shipped a new release, validated against the latest stable Kubernetes and hardened with stricter default security policies for multi-tenant clusters. No date has been set for a CNCF Graduation vote, but a new Outreach Program and Contributor of the Month effort are meant to widen Kubeflow's contributor base ahead of the review, and the community is planning a half-day virtual event on August 19 focused on how the platform is being used for generative AI and MLOps in production.
