ai-agents-llm-thinking-mode-di-deepseek-models-return-empty-or-severely-truncated-e7a9cf60

THEN Add a per-provider extra_body field in the Manifest provider configuration with {"thinking": {"type": "disabled"}} to disable DeepSeek thinking mode. This can be implemented as a provider-level setting or a per-model override in routing tiers.

{
  "model": "deepseek-chat",
  "messages": [...],
  "thinking": {"type": "disabled"}
}