Baseten announced plans for three additions to its inference platform across Nov. 4–6, 2026. In a Nov. 3 X article, @amiruci set out the schedule: a Grid public preview on Wednesday, Sandboxes on Thursday and Loops general availability on Friday. Those were planned rollouts, not a report that the launches had happened.
What each rollout is meant to do
Grid is Baseten's proposed control layer for coding agents. Its gateway is designed to bring models and spending into one place, an intelligent router would match tasks to models, and guardrails would set company-wide policies for model use. Baseten planned to open Grid as a public preview on Nov. 4.
Baseten describes Sandboxes as stateful, pause-and-resume environments isolated in virtual machines. The company claims they can scale to thousands of instances in seconds and plans to place them alongside inference. Its target launch date was Nov. 5; the scaling claim is Baseten's, not a measured result presented here.
Loops handles serverless training on open-weight models, including supervised fine-tuning and reinforcement learning. Baseten said Loops was already in public preview and planned to make it generally available Nov. 6. The company also describes an integration with Dedicated Inference that lets customers serve a trained model as a production endpoint. November's planned change was general availability, not Loops' first introduction.
