New Display Field for Extended Thinking

Anthropic has launched a display field for Claude's extended thinking capability, giving developers finer control over how thinking content appears in API responses. This addition addresses a key developer need: streaming responses faster when the intermediate reasoning process isn't needed in the final output.

How It Works

Developers can now set thinking.display: "omitted" in their requests to suppress thinking content from response streaming. When configured this way, Claude will:

• Return thinking blocks with an empty thinking field (the actual reasoning is computed but not transmitted)
• Preserve the signature field for multi-turn conversations that rely on thinking continuity
• Maintain the same billing structure as before

Use Cases and Benefits

This feature is particularly valuable for applications where:

• Latency matters: Omitting thinking blocks reduces the amount of data streamed to users
• Cost optimization: Developers can leverage extended thinking's reasoning capabilities without paying for transmission of thinking tokens
• Multi-turn conversations: The preserved signature ensures subsequent turns can build on the model's reasoning even when the thinking content isn't displayed

Developers can learn more and see implementation examples in the Controlling thinking display documentation.

Context

This update is part of Anthropic's broader extended thinking initiative, which allows Claude models to spend more computational effort on complex reasoning before responding. The new display field complements existing thinking controls and provides developers with more granular options for tuning performance and costs.