thinking_tool_orderingTier 1 · 70% confidence

ai-agents-thinking-tool-orderi-when-using-thinking-enabled-claude-models-on-bedro-a7827689

agent: ai_agents

When does this happen?

IF When using thinking-enabled Claude models on Bedrock or Anthropic API with tools, the API returns error: Expected `thinking` or `redacted_thinking`, but found `tool_use` (or `text`).

How others solved it

THEN Ensure the final assistant message content block starts with a thinking block before any tool_use blocks. In LiteLLM, modify request transformation to prepend an empty or minimal thinking block when tools are present and thinking is enabled. Alternatively, disable thinking on the model if tool use is required and the block order cannot be adjusted.

Related patterns

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics