Skip to content
context Coding

Beta test of per-message effort changes on Claude API preserves the prompt cache when model effort changes

clearly official source

In a beta test on Claude API, Anthropic allows the model effort level to be changed in individual conversation messages without losing the prompt cache, for the models Claude Fable 5.1, Mythos 5.1 and Opus 5.

Anthropic has launched a beta test of per-message effort changes on Claude API. The feature allows the model effort level to be changed for individual messages within an ongoing conversation without losing the prompt cache already created. The beta is available for the models Claude Fable 5.1, Claude Mythos 5.1 and Claude Opus 5.

According to Anthropic, the feature is activated by adding a message with the role "system" containing the output_config.effort field to the conversation messages — the setting then applies to subsequent turns. Using it requires including the beta header mid-conversation-output-config-2026-07-01 in API requests.

The source text does not specify a production release date, specific pricing terms or further technical details. Details can be found in the source article.

What changed

Why it matters

Until now, changing the model effort level in the middle of a conversation usually required rebuilding the context, thereby losing the benefits of the prompt cache (lower costs and latency for reused context). The new beta feature removes this limitation, which is particularly relevant for developers of agentic applications, where the need for computing power varies from query to query.

Two audiences, two different impacts

What this means

01

For individuals

Developers working with Claude API can now change the model effort level for individual messages in a conversation without invalidating the prompt cache, allowing them to tailor the model response to the needs of a specific query without incurring caching costs again.

What to do Try the beta feature output_config.effort with the mid-conversation-output-config-2026-07-01 header in a test API call.
More practical updates →
02

For a business

Companies running products on Claude API may save on the costs and latency associated with rebuilding the prompt cache when dynamically adjusting model effort within a conversation.

Development
What to decide Evaluate whether using per-message effort changes in products built on Claude API will reduce the costs of caching prompts again.
More business impacts →
Claude API Claude Fable 5.1 Claude Mythos 5.1 Claude Opus 5 per-message effort

Check the original

Event sources

clearly official source · 1 publisher, 0 independent. We count feeds from the same owner only once.

1
Claude Platform Release Notes primary source · first detected Per-message effort changes are in beta on Claude Fable 5.1, Claude Mythos 5.1, and Claude Opus 5 on the Claude API. Add a role: "system" message with output_config.effort inside…