OpenAI donates API standard designed to ease switching between AI providers

OpenAI donated Open Responses, its specification for a common interface to AI models, to the Agentic AI Foundation on October 22, 2026. The move puts a project intended to make switching model providers easier under foundation stewardship. It is one of four projects accepted in the same announcement, bringing the foundation’s portfolio to 10.

The other contributions came from Anthropic, Hugging Face and Alibaba.

A common interface, not a common model

Developers connect applications to cloud-based language models through application programming interfaces, or APIs. Different providers use different interfaces, so software written for one may need extensive changes before it works with another. Open Responses supplies a standardized template for those interfaces. Its purpose is to let developers move code between APIs that follow the specification without major rewrites.

The mismatch is not necessarily about entirely different capabilities. When OpenAI introduced the specification in January, the project team pointed to shared building blocks such as messages, tool calls and function calling. Providers offered those elements but represented them differently in their APIs.

OpenAI modeled the specification on its own model-access API. That gives developers already familiar with OpenAI’s interface a starting point, and is intended to ease movement between OpenAI workloads and compatible APIs.

10Projects under stewardship

The four donations bring the foundation to 10 projects, compared with three when the Linux Foundation launched it the previous December.

Instructions and environments tackle different handoffs

Two of the other donations focus on sandboxes: environments that simulate the infrastructure a model will use in production. For a coding model, that might mean a simulated developer workstation.

• Anthropic’s Agent Skills provides a standard format for bundling instructions, examples and automation scripts for AI agents. Anthropic released the open-source project the previous December. Consistent packaging is intended to make those resources easier to share between development teams, without restructuring the files before connecting them to another workload.
• Hugging Face’s OpenEnv addresses switching between training systems. Those systems coordinate sandboxes and automate repetitive management work during AI training. OpenEnv is intended to reduce the effort required to change the system doing that coordination.
• Alibaba’s OpenSandbox supplies APIs for managing the sandboxes themselves. Its tasks include connecting agents running inside those environments to external applications.

For Anthropic, the donation follows an earlier transfer to the same organization. Engineer David Soria Parra said the foundation had given Model Context Protocol, or MCP, a “neutral home” since Anthropic donated it less than a year ago. He also said Anthropic remained committed to investing in open-source agent tools alongside the foundation.