OpenAI Responses API
A next-generation API specification introduced by OpenAI as a replacement for the Chat Completions API. Core features include persistent conversation state, native tool-calling support, and an event-stream response structure. The API has been adopted by OpenAI's own tools, including Codex CLI.
Timeline (last 90 days)
For the same MiniMax M3.1 Flash model, the effort field is named reasoning_effort, output_config.effort, and reasoning.effort under the OpenAI Chat Completions, Anthropic Messages, and OpenAI Responses protocols respectively; all three accept the five values low/medium/high/xhigh/max, and all default to max.
Setting api: "responses" in the Mastra model configuration causes the agent to instead POST to /v1/responses and use an input array in place of messages.
PR #18911 allows Ollama's OpenAI-compatible layer to accept and pass through custom tool calls in Responses history.
The problems caused by output_index drift may manifest as truncated output content, garbled message block concatenation, or in some cases trigger parsing exceptions.
langchain-openai 1.7.2 fixes the issue in the OpenAI Responses API where, when output_index drifts in streaming scenarios, streamed return items are incorrectly split across multiple blocks (PR #41121).
The OpenAI Responses API introduces a session state persistence mechanism, allowing the server to save conversation context so that clients no longer need to include the full message history in every request, thereby reducing token consumption.
DeepSeek V4 Pro为Flash版本引入了灵活的推理努力（reasoning effort）选项，并新增对OpenAI Responses API的支持
All Facts (7)
PR #18911 allows Ollama's OpenAI-compatible layer to accept and pass through custom tool calls in Responses history.
70%UnverifiedFor the same MiniMax M3.1 Flash model, the effort field is named reasoning_effort, output_config.effort, and reasoning.effort under the OpenAI Chat Completions, Anthropic Messages, and OpenAI Responses protocols respectively; all three accept the five values low/medium/high/xhigh/max, and all default to max.
50%UnverifiedSetting api: "responses" in the Mastra model configuration causes the agent to instead POST to /v1/responses and use an input array in place of messages.
50%UnverifiedThe problems caused by output_index drift may manifest as truncated output content, garbled message block concatenation, or in some cases trigger parsing exceptions.
50%Unverifiedlangchain-openai 1.7.2 fixes the issue in the OpenAI Responses API where, when output_index drifts in streaming scenarios, streamed return items are incorrectly split across multiple blocks (PR #41121).
50%UnverifiedThe OpenAI Responses API introduces a session state persistence mechanism, allowing the server to save conversation context so that clients no longer need to include the full message history in every request, thereby reducing token consumption.
50%UnverifiedDeepSeek V4 Pro为Flash版本引入了灵活的推理努力（reasoning effort）选项，并新增对OpenAI Responses API的支持
50%