Extended Thinking Display Control

Anthropic has launched the display field for extended thinking on the Claude API, giving developers fine-grained control over how thinking blocks are returned in responses. By setting thinking.display: "omitted", developers can now receive thinking blocks with an empty thinking field while preserving the signature for multi-turn conversation continuity.

Key Benefits

• Faster streaming: Omitting thinking content from responses enables faster streaming without including verbose intermediate reasoning
• Maintained continuity: The preserved signature allows the model to reference previous thinking blocks across multiple turns
• No billing impact: This feature does not change pricing—developers are still billed the same way regardless of display settings
• Backwards compatible: The feature works alongside existing thinking functionality

Implementation

Developers can control thinking display in their API requests via the configuration parameter. Full documentation is available in the Controlling thinking display guide on the Claude Developer Platform.

Context

This release is part of a larger March 2026 update to the Claude API that also includes the general availability of the 1M token context window for Claude Opus 4.6 and Sonnet 4.6 at standard pricing, an increase in media limits from 100 to 600 images or PDF pages per request, and other enhancements to extended thinking capabilities.