Open Responses: A New Standard for AI Agent Inference

Open Responses: What you need to know

Open Responses is a new, open inference standard built by the open source AI community, backed by the Hugging Face ecosystem, and initiated by OpenAI. It’s based on the Responses API and designed for the evolving needs of AI Agents, overcoming the limitations of the older Chat Completion format.

The shift to autonomous agents demands more than turn-based conversation support; yet, the Chat Completion format remains prevalent despite its inadequacy for complex, agentic tasks. This introduces inefficiencies and necessitates a standardized approach for advanced AI systems.

Why This Matters

Currently, developers often rely on workarounds and undocumented extensions to the legacy Completions API, leading to inconsistent and fragile inference experiences. Replacing these with a formalized, open standard reduces integration costs and improves code maintainability – especially given that agentic workflows are rapidly increasing in complexity and scale.

Key Insights

• Responses API launched by OpenAI, 2025: Marked the initial effort towards addressing limitations of previous APIs.
• Stateless design with optional encryption: Allows flexibility for providers, accommodating both performance and privacy requirements.
• Semantic streaming: Provides granular updates during inference, improving responsiveness and observability.

Working Example

curl https://evalstate-openresponses.hf.space/v1/responses \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $HF_TOKEN" \
-H "OpenResponses-Version: latest" \
-N \
-d '{
"model": "moonshotai/Kimi-K2-Thinking:nebius",
"input": "explain the theory of life"
}'

Practical Applications

• Hugging Face Inference Providers: Implementing Open Responses to offer a standardized interface for accessing various models.
• Agent Development: Facilitating the creation of robust and scalable agents with enhanced reasoning and tool usage capabilities.

References:

• https://huggingface.co/blog/open-responses