Problem

When using Xiaomi MiMo models (e.g., mimo-v2.5-pro) via the xiaomi provider, Hermes does not pass the thinking parameter in the API request.

According to MiMo official documentation, thinking mode is enabled by default, which significantly increases token consumption. Even when the user configures thinking: hidden or reasoning_effort: none in config.yaml, these settings are not forwarded to the MiMo API endpoint.

Expected Behavior

Hermes should respect the user's thinking / reasoning_effort configuration and pass the appropriate parameter (e.g., thinking: { type: "disabled" } or equivalent) to the MiMo API when the user explicitly disables thinking.

Current Behavior

• The thinking parameter is not included in API requests to MiMo
• MiMo defaults to thinking=enabled, consuming extra tokens
• Config settings like thinking: hidden and reasoning_effort: none have no effect on the actual API call

Config Example

model:
  provider: xiaomi
  default: mimo-v2.5-pro
  api_key: tp-xxx

agent:
  reasoning_effort: none

display:
  sections:
    thinking: hidden

Impact

Unnecessary token consumption on every request when using MiMo models, especially costly for users on limited token plans.

Environment

• Hermes Agent (latest)
• Provider: xiaomi
• Model: mimo-v2.5-pro
• API endpoint: https://api.xiaomimimo.com/v1