How Baseten announced its AI system building blocks on X
Inference platform for serving and training custom AI models in production
What Baseten launched
Inference, training, routing, governance, and sandboxes are the building blocks of every AI product today. We’ve learned a lot from our customers about what it takes to build that system well: @OpenEvidence trains models that outperform the closed frontier, then serves them on https://t.co/z8Bhe2FLpj
How TLVC rated the launch
TLVC Rating
About the launch
Baseten provides inference infrastructure for product and platform teams that serve open-source, private, and custom models in production, on dedicated GPU deployments or through hosted model APIs. This launch adds routing, governance, agent execution, and post-training around that serving. Grid controls which models coding agents can call and what they cost. Sandboxes give agents an isolated place to run code. Loops, reaching general availability the same week, is the training SDK for supervised fine-tuning and reinforcement learning, with a checkpoint that can deploy back to a dedicated inference endpoint. OpenEvidence trains custom models on Baseten and serves them on dedicated inference, Datadog post-trains on Loops, and Opendoor routes company AI requests through Grid. Grid is for organizations that want central control over coding-agent models and spend. Admins approve open-weight models from Baseten Model APIs, or Anthropic, OpenAI, and xAI models billed to the company's own keys, then set a monthly budget for each member and for the team. A direct route always calls one model. An auto route uses a managed classifier to choose among approved models for each request. Developers stay in Claude Code, Codex, OpenCode, or Pi, and a setup command points those tools at the team's routes. Admins can review spend, tokens, and cache hit rate by team, developer, model, and provider. Grid opens as a public preview. Sandboxes launch the same week as isolated environments where an agent can run commands and work with files, one per user or project. They are stateful, VM-isolated, and co-located with inference, can pause and resume with files and processes intact, and are built to scale to thousands of instances in seconds. The product builds on Blaxel, which Baseten acquired in September 2026. Loops is a Python SDK for LoRA supervised fine-tuning and reinforcement learning on open-weight models. The customer writes the data, reward functions, and training loop, while Baseten runs the trainer and sampler. A finished checkpoint can deploy to a dedicated inference endpoint.
Tags
Questions
Frequently asked questions
The Baseten launch post reached 36K views on X, with 156 likes, 20 reposts, 17 comments, 70 bookmarks.
Baseten launched on X on November 3, 2026 with the message "Inference platform for serving and training custom AI models in production".
Baseten provides inference infrastructure for product and platform teams that serve open-source, private, and custom models in production, on dedicated GPU deployments or through hosted model APIs.
Launchvideo.com rated the Baseten launch 2.80 out of 5: hook 1/5, editing / creativity 4/5, copy 4/5, sentiment of launch 4/5, distribution strategy 1/5.
