Baseten brings frontier open models to Microsoft Foundry

Baseten’s frontier open models are now available in public preview through Microsoft Foundry, giving Azure customers a first-party Microsoft service for testing, deploying and scaling them. The models are powered by Baseten’s inference infrastructure and can be selected from the Foundry model catalog.
- Baseten is available in public preview on Microsoft Foundry.
- Launch models include DeepSeek V4.1 Flash, GLM-5.3, GLM-5.3-Flash and Kimi K3.
- Customers can choose Standard or Provisioned deployments.
- Requests are stateless, and Baseten stores no prompts or responses.
- General availability is expected in early 2027.
Open models join Microsoft Foundry’s catalog
The partnership gives organizations another option alongside the open and closed models already available in Microsoft Foundry. According to Baseten, teams can evaluate models for coding agents, research workflows and customer support applications, then put them into production on the same platform used to build and govern AI applications and agents.
The launch models are DeepSeek V4.1 Flash, GLM-5.3, GLM-5.3-Flash and Kimi K3. Baseten says additional models will become available on the day they are released. The company’s service handles model serving infrastructure, allowing teams to test open models without sourcing GPU capacity or managing model runtimes.
Microsoft Foundry customers can use the models under the same enterprise controls as other models in the platform. Usage is billed through Azure as first-party Foundry consumption and counts 100% toward a customer’s Microsoft Azure Consumption Commitment, or MACC.
Deployment and enterprise controls
Baseten says its performance work covers every layer of the inference stack, including routing, batching, caching and kernels. The infrastructure is designed for 99.9% uptime, according to the company.
Customers can choose between Standard and Provisioned deployments. Global Standard and Data Zone Standard deployments use pay-per-token billing, while Global Provisioned and Data Zone Provisioned deployments use provisioned throughput units, or PTUs, for reserved capacity and predictable throughput.
The integration also includes zero data retention with stateless requests. Microsoft Entra ID and key-based authentication are available through Foundry, along with per-model access controls using Azure Policy. Baseten says it stores no prompts or responses.
“Bringing Baseten to Microsoft Foundry gives enterprises more choice and pairs Baseten’s optimized and reliable inference with the governance, controls and observability within the Foundry platform,” Steve Sweetman, vice president of product at Microsoft Foundry, said in the announcement.
How Azure teams can access the models
Baseten models are available to all customers in the Foundry model catalog, but an administrator must opt in to enable access for an organization. Developers can then select a Standard or Provisioned deployment, with deployments ready in seconds, according to Baseten.
The models are served through the Chat Completions API and the Responses API. Baseten says this allows existing OpenAI-compatible code and agent workflows to carry over with minimal changes.
The public preview is the first phase of the offering. Baseten expects general availability in early 2027 and says it will continue adding frontier open models as they are released. For Azure enterprises, the announcement expands model choice while keeping deployment, governance, observability and billing within Microsoft Foundry.
Baseten: Baseten is now available on Microsoft Foundry


