Moonshot AI's open model Kimi K3 is now available to enterprise users through OpenAI's Codex channel in an arrangement facilitated by United States-based inference provider Baseten. The integration allows OpenAI enterprise customers to apply their existing spending commitments to usage of the Chinese open model without signing a separate supplier contract. Sina Finance reported that this marks the first time a Chinese open model has been included in OpenAI's enterprise billing system.
Partnership Details
Baseten described the collaboration as one of the first open-model inference partnerships in the OpenAI B2B Marketplace, with native integration inside Codex. Enterprise customers can access open models served by Baseten directly within Codex or through the Responses API. In this setup, Baseten provides model inference while Codex serves as the developer interface, allowing organizations to use existing OpenAI commitments.
Expanding Multi-Model Coding
Agentic coding has become one of the most widely adopted enterprise artificial intelligence use cases, and OpenAI's Codex and GPT models are popular tools for scaling code generation workflows. Baseten argues that leading companies increasingly use a mix of open and closed models to balance intelligence, cost, latency, and capacity. The new integration is intended to help organizations route each coding task to the model best suited for that specific job.
Inference Performance and Reliability
Code generation workloads create demanding inference conditions because of multiple turns, long prompts over large repositories, and extensive context windows. Baseten says its platform spans more than 90 clusters across over 20 clouds, with active-active deployments that allow traffic to be rerouted if a provider or region fails. The company also maintains relationships with more than 200 compute providers to keep capacity ahead of enterprise demand.
Developer Tools and Engineering Support
Baseten has also expanded beyond inference by incorporating Blaxel, which provides sandboxes where agents can run the code they write. Forward-deployed engineers help tune deployments to specific latency targets, while applied researchers support post-training work with customers. The company highlights its work with LangChain to train custom models for LangSmith Engine using Baseten Loops as an example of this support model.
Governance and Day Zero Access
Baseten promotes day zero access to major open models, meaning new frontier releases can be evaluated on the day they ship without waiting for lengthy procurement cycles. The platform runs on United States-based infrastructure with zero data retention for prompts, and customers can pin deployments to specific regions for stricter compliance requirements. Fine-grained authentication, authorization controls, and usage visibility across models, users, and keys are also available.
Strategic Significance
The arrangement is notable because it places a Chinese open model inside OpenAI's enterprise billing environment, a development that Sina Finance framed as unprecedented. It also illustrates how quickly enterprise artificial intelligence adoption is moving toward a multi-model future, even across geopolitical boundaries in the AI supply chain. For enterprise buyers, the immediate benefit is simplified procurement and the ability to test a competitive model without new vendor agreements.
The Baseten and OpenAI partnership represents a practical step toward multi-model agentic coding workflows in enterprise environments. By allowing existing OpenAI commitments to cover Kimi K3 usage, the arrangement lowers administrative barriers and gives customers more flexibility in model selection. It also signals that open model inference providers are becoming a deeper part of mainstream enterprise AI procurement and deployment.