MiniMax 把 OpenAI 新 Agents API 做成开源版，Codex、Claude Code 都能接 --- 这个消息挺有意思的，几个点值得拆开看： **1. 时机选得准** OpenAI 刚发布 Agents API（Responses API + 内置工具链），生态还没完全铺开，MiniMax 就推出了兼容的开源替代方案。这是典型的"趁热打铁"策略——开发者对 Agents 能力有需求，但可能不想被 OpenAI 绑定，或者想本地部署/自托管。 **2. "都能接"是关键卖点** Codex、Claude Code 这些工具本身是各家自己的 Agent 框架，MiniMax 做的是**协议兼容层**——让这些工具能通过统一的 API 格式调用 MiniMax 的模型和 Agent 能力。本质上是在做"Agent 领域的 OpenAI 兼容层"，类似当年 vLLM/Ollama 做推理兼容那样。 **3. 开源版的定位** 如果确实是开源，那大概率是： - 提供兼容 OpenAI Agents API 的 server 实现 - 后端可接 MiniMax

动察 Beating AI News Flash: MiniMax has open-sourced OpenAgentCore, a self-hosted implementation compatible with the OpenAI Agents API. Developers can continue using the official OpenAI SDK, only needing to swap the API address, and the Agent running in the backend can choose Codex, Claude Code, or MiniMax Code.

The Agents API can be understood as a "unified interface for calling Agents." A regular model API is mainly responsible for one question and one answer. An Agent needs to work continuously for hours or even days, and also manage context, call tools, coordinate sub-agents, and save task progress. OpenAI's Agents API puts these capabilities into a long-running Session for unified management.

The official OpenAI version uses OpenAI's hosted Codex execution framework. Developers can choose where the code runs themselves, but the Agent's task orchestration, context management, and runtime state are still handled by OpenAI. OpenAgentCore, by contrast, also lets developers deploy this backend themselves.