Baseten has brought a selection of open AI models to Microsoft Foundry in public preview, giving Azure customers another way to try and deploy them within their existing enterprise setup. The practical draw is a mix of model choice, Microsoft’s controls and familiar billing, rather than another model catalogue with a fresh login to remember.
Baseten Watch analysis
What happened
The models are available in the Foundry model catalogue. Baseten names DeepSeek V4.1 Flash, GLM-5.3, GLM-5.3 Flash and Kimi K3 among the launch selection, and says more models will follow as they are released. An administrator must opt in for an organisation before developers can use them.
Customers can choose Standard deployments, billed per token, or Provisioned deployments using reserved throughput. The models work with Foundry’s Chat Completions and Responses APIs, which Baseten says lets existing OpenAI-compatible code and agent workflows carry over with minimal changes. The company’s announcement of the Foundry preview also says requests are stateless and Baseten stores no prompts or responses.
Our top picks
- Open models alongside existing Foundry models
Teams can compare open options with the models they already use on the same platform. - Two deployment routes
Standard offers pay-per-token usage; Provisioned reserves throughput for more predictable capacity. - Enterprise access controls
The announcement lists Microsoft Entra ID, key-based authentication and per-model access controls through Azure Policy. - Familiar APIs
Chat Completions and Responses support may make it easier to adapt existing OpenAI-compatible workflows. - Azure billing and commitment eligibility
Baseten says usage is billed as Foundry consumption and counts towards Microsoft Azure Consumption Commitment spending.
Why it matters
Enterprises weighing open models often face a less glamorous hurdle than model quality: getting infrastructure, governance and procurement to agree to share a room. This preview puts Baseten-powered models inside Foundry’s existing controls and billing, while offering a choice between usage-based and reserved capacity. That could make testing easier for organisations already invested in Azure.
The announcement is from Baseten, which describes the service’s performance and reliability in its own terms. Public preview is available to Foundry customers today, but general availability is expected in early 2027. For now, an admin opt-in is part of the trip.
Our read
This is a useful distribution move for open models, and the details matter more than the partnership headline: named models, selectable deployment types and support for familiar APIs give teams something concrete to evaluate. Azure customers should compare actual model performance, latency and costs on their own workloads before treating the catalogue as a bargain shelf. The sticker is not the bill, as cloud users know rather too well.
What to watch
- Which models Baseten adds, and how quickly they appear in Foundry.
- Whether independent customer testing supports the company’s performance and reliability claims.
- When the service reaches general availability, expected in early 2027.
Discussion spark: For enterprises already on Azure, do familiar governance and billing make open models meaningfully easier to adopt, or will model performance still decide everything?
Sources and evidence
- Baseten is now available on Microsoft Foundry (17 November 2026, 21:22 UTC)
Independent WittyWires tracker for public updates about Baseten. Not affiliated with or endorsed by Baseten; this is not an official account.