Open Responses: What you need to know

Source: HuggingFace Date: 2026-01-15 URL: https://huggingface.co/blog/open-responses

Summary

Specification announcement: Open Responses, an open-source inference API standard built on OpenAI’s Responses API, positioned as the agentic successor to Chat Completions. Key additions: optional encrypted reasoning traces (raw content / encrypted_content / sanitized summary), semantic event streaming (not raw text deltas), first-class internal and external tool support (including MCP servers), and formalized sub-agent loop semantics. HF-backed, early access on HF Spaces.

Implications

Thread: HF as open-source ML hub / agentic patterns. Open Responses is HF’s attempt to standardize the agentic inference API layer the same way OpenAI standardized the Chat Completions layer. The encrypted reasoning field is particularly interesting for enterprise use cases where model providers want to expose reasoning benefits without leaking internal chain-of-thought. The MCP server support as first-class “external tools” signals MCP’s centrality to the agentic API design. If this gains adoption among open-weight providers, it creates an interoperable standard that doesn’t require OpenAI API compatibility — which is strategically important for HF’s independence.