Anthropic issued guidance after launching Claude Opus 5.5 that asks developers to retest effort settings, remove some ‘think carefully’ chat-lines, use time budgets for agents, and mark pasted external text. These are practical, model-specific tuning suggestions—not a requirement to rewrite all prompts.
Remove chat-system lines that tell the model to "think carefully"—Opus 5.5 chooses its own thinking level and removing that line sped replies without clear quality loss in tests.
For agents, pass elapsed time as a time budget and tag pasted external text with IDs and handling instructions as a guardrail against prompt injection.
# What changed with Claude Opus 5.5
Opus 5.5 runs at medium effort by default. Opus 5 used a higher default. Anthropic recommends that effort be the first lever you change when you want different speed, cost, or quality tradeoffs. In their internal tests, Opus 5.5 at medium effort matches or beats Opus 5 at high effort on coding and knowledge tasks.
Practical step: run the same prompts you used on Opus 5 against Opus 5.5 at multiple effort levels (lower, medium, higher) and measure latency, cost, and quality rather than assuming prior settings remain optimal.
Practical step: for chat UIs, strip lines that explicitly instruct the model to deliberate and test response time and quality with and without them.
If you want the model to think less, try lowering the effort setting first rather than adding prompt instructions that ask for less thinking. Save very high effort like xhigh or max for tasks where you can demonstrate a real quality gain.
Practical step: build effort as a tunable parameter in your integration and make it easy to test different levels in QA.
Practical step: implement a task-level time budget and include elapsed-time metadata so agents prioritize within the budget.
When users paste emails or other source text, Anthropic suggests wrapping that content with tags containing a random ID and a system note describing how Claude should treat the tagged text. Tags are plain text and copyable, so this is only one layer of defense against prompt injection.
Practical step: add textual tags plus handling notes for pasted content and treat them as a partial guardrail.
Anthropic has released Claude Fable 5.1 and Mythos 5.1, addressing issues such as performance, data retention, safeguards and price. The updates also target scientific discovery, as the company tested the capabilities of the models across many domains. Internal testing shows that Fable 5.1 outperforms Fable 5 when is c
Anthropic just updated its frontier AI model, Opus. The new model, Opus 5.5, demonstrates even better results than its predecessor, Opus 5. Continue reading on UX Planet »
Anthropic today launched Claude Opus 5.5 , which it says is the first model in a new 5.5 model family. Opus 5.5 performs at the level of Claude Fable 5.1 on most work, but costs 40 percent less to run than Opus 5. According to Anthropic, Opus 5.5 is a "major step up" from Opus 5 with improvements in performance and saf
Anthropic today introduced Claude Opus 5.5, the first release in a new family of Claude 5.5 models. Claude Opus 5.5 performs at the level of Claude Fable 5.1 on most work, and costs around 40% less to run than Opus 5. Over the coming weeks, we’ll also be launching Claude Sonnet 5.5 and Haiku 5.5.... … continue reading
Anthropic said Opus 5.5 costs 20% less than Opus 5 and is available on AWS, Google Cloud and Azure
Anthropic says Claude Opus 5.5 rivals Fable 5.1 on most work at 40 percent lower cost than Opus 5, with tougher safeguards.
Loading more related stories...
Open the app view to save this story, compare related coverage, and continue from the same source.