XiaoMi Mimo

[jiangchuan2]

[English / 中文]

Issue: MiMo model returns 400 error due to missing reasoning_content in multi-turn tool calls
问题：MiMo 模型在多轮工具调用时因缺少 reasoning_content 返回 400 错误

What happens / 现象
When using Xiaomi MiMo model (e.g. mimo-v2.5-pro) with reasoning mode enabled, if a conversation contains tool calls, the API requires that every subsequent assistant message must include the full reasoning_content from the previous turn. Otherwise it returns HTTP 400.
使用小米 MiMo 模型（如 mimo-v2.5-pro）并开启思考模式时，如果对话中发生过工具调用，后续每一轮的 assistant 消息必须原样回传上一轮的 reasoning_content 字段，否则 API 返回 400。

Current behavior in OpenClaw / 当前行为
OpenClaw does not store or replay reasoning_content. When a conversation has multiple turns + tool calls, it fails with 400.
OpenClaw 没有保存和回传 reasoning_content。一旦对话出现多轮 + 工具调用，就会触发 400 错误。

Expected behavior / 期望行为
OpenClaw should automatically cache reasoning_content from assistant responses and reinject it in subsequent requests, similar to how tool_calls is handled.
OpenClaw 应自动缓存 assistant 回复中的 reasoning_content，并在后续请求中自动补回，类似目前对 tool_calls 的处理。

Workaround / 临时方案
Setting "reasoning": false in model config avoids the error but disables reasoning mode, reducing quality for complex tasks.
在模型配置中设置 "reasoning": false 可绕过，但会关闭思考模式，降低复杂任务的质量。

Environment / 环境

• OpenClaw version: 2026.4.23
• Model: Xiaomi MiMo mimo-v2.5-pro (Max subscription)
• API: OpenAI-compatible (/v1/chat/completions)

Request / 请求
Please add native support for storing and replaying reasoning_content.
请官方增加对 reasoning_content 的存储与回传支持。

---

直接复制上面整段就行，中英文都在一起，国内国外的人都能看懂。