ai-agents-llm-routing-config-deepseek-models-v4-flash-v4-pro-return-empty-or-se-171c96d4

THEN Remove DeepSeek models from all routing tiers in Manifest, or configure the provider to pass extra_body with {"thinking": {"type": "disabled"}} to disable thinking mode. If removal is not feasible, set max_tokens to at least 4000 to avoid content truncation.

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