From the source
Baseten's inference infrastructure is in public preview through
Microsoft Foundry, initially serving four open models to Azure customers as a first-party Microsoft service.
- The lineup is
DeepSeek V4.1 Flash, GLM-5.3, GLM-5.3 Flash and Kimi K3, sitting in the Foundry catalog under existing enterprise controls and customers' Microsoft agreements.
- All Foundry customers can use the preview, but an administrator must opt in for their organization.
- Baseten says it handles serving so teams can evaluate open models without sourcing GPU capacity or managing runtimes, with Standard and Provisioned deployments ready in seconds.
- Models are served through the Chat Completions and Responses APIs, which Baseten says lets existing
OpenAI-compatible code and agent workflows carry over with minimal changes.
- Requests are stateless and Baseten stores neither prompts nor responses; usage is billed through Azure as first-party Foundry consumption, and Baseten says 100% of that spending counts toward a customer's Microsoft Azure Consumption Commitment.
- Baseten expects general availability in early 2027 and plans to add open models on their release day.