OpenAI gives its Open Responses standard to a neutral group
Many providers use the standard, but Anthropic and Google use their formats.
On Thursday, OpenAI gave the Open Responses specification to the Agentic AI Foundation, a part of the Linux Foundation. The specification is one set of API calls for many model providers. AWS, Databricks, Nvidia and Hugging Face use it. The foundation launched in December 2025 and gives the project a neutral home. OpenAI gave no test results.
How it works
Providers show the same parts, such as messages and tool calls (a task that a model tells software to do), in different formats. Open Responses sets one format for these parts. A developer can change models and make less changes to the code.
Who uses it
OpenAI says Vercel AI Gateway and Databricks Model Serving use it. Ollama and vLLM also use it. The project says that its primary schema files come from the OpenAI API.
What is not known
Anthropic is a member of the foundation but has a different Messages API. Google has an Interactions API. DeepSeek, Moonshot, Z.ai and MiniMax give endpoints for the Anthropic format. It is not known if Anthropic or Google will use Open Responses.
Insight: Open standards now compete for the format
In 2026, Meta and Sierra started an open protocol for agents of one person. Also, NVIDIA made its platform for GPU use open source. OpenAI gives a model API format. This seems to be a move to control the format, because the format that most tools copy controls all other tools. The Anthropic format is the other large format. Monitor which one the next providers copy.
Sources
Posted
