OpenAI gave away the API spec developers were already using

OpenAI has handed over Open Responses, the specification it wrote for a common LLM API, to the Agentic AI Foundation. The New Stack reports that the spec joins MCP, A2A and OpenAI's own AGENTS.md in the foundation, which sits inside the Linux Foundation. Anthropic's Agent Skills and Hugging Face's OpenEnv framework for agent environments are being contributed at the same time.

Writing for The New Stack, Frederic Lardinois quotes an OpenAI spokesperson: "When we co-founded AAIF, we contributed AGENTS.md, a way to give coding agents project-specific instructions. Open Responses addresses another part of building agents by giving developers a shared format for working with models across endpoints that support it, so they have fewer provider-specific integrations to maintain. Vercel AI Gateway and Databricks Model Serving already implement it, and AAIF will provide a neutral home for developers and providers to continue developing the specification together."

What the spec does

Open Responses is a shared set of core API calls that should work across LLM providers. When OpenAI first announced the specification earlier this year, The New Stack writes, the team pointed out that every LLM provider offers the same building blocks, messages, tool calls, function calling, but each represents them differently in its own API. The spec standardises those primitives so a developer swapping between models does not have to translate between each vendor's version of the same idea.

The project is open about its lineage. The contributor guide notes that the core schema files are copied from OpenAI's own API, and OpenAI's earlier Chat Completions API became an early standard that other providers modelled their offerings after.

Who signed up

The New Stack lists wide support for the spec from AWS in Bedrock, Databricks, LM Studio, Ollama, Hugging Face, Nvidia, OpenRouter, Red Hat and Vercel, with the open-source projects vLLM and OGC, previously known as Llama Stack, backing it as well.

Among the names missing from the list is Anthropic, which defines its own Messages API for its models. Anthropic is a member of the AAIF, so Lardinois notes the company may adopt the spec at some point in the future. Google also has its own default in the Interactions API. Both are large enough that a developer switching between every major provider still has at least two API shapes to deal with today.

Lardinois writes that the Messages API has left a clear mark thanks to the success of Claude Code, with DeepSeek, Moonshot, Z.ai and MiniMax exposing Anthropic-compatible endpoints. That split is the fragmentation a vendor-neutral spec like Open Responses is meant to fix.

What a neutral home changes

Companies donate projects to a foundation to put them under shared governance. The Linux Foundation has become the default place for that, with a technical steering committee of project maintainers handling the technical discussions. The donation moves Open Responses off a single company's roadmap and onto one every member has a say in.

The AAIF now sits in a comparable spot to the CNCF for cloud-native work. It holds MCP, A2A for agent-to-agent messaging, AGENTS.md, the open-source goose agent, Agent Router, agentgateway, OpenSandbox and the two new additions. The API call between a user and a model is now one layer in a growing stack of specs for building agents.

Source

• OpenAI donates Open Responses, its spec for a common LLM API, to the Agentic AI Foundation by Frederic Lardinois, The New Stack
• Open Responses project page at the Agentic AI Foundation