A year and a half ago, users could sometimes coax more performance out of a model or steer the router this way, whereas Claude Opus 5.5 now scales its reasoning depth differently. According to Anthropic, outdated prompting habits carried over from older model generations are slowing down Claude Opus 5.5. The model has become the default on Pro, Max, Ultra, and Team plans, yet users often hold it back with outdated instructions.
This gap between model architecture and legacy prompting translates into unnecessary latency and wasted token spend. As Anthropic stated, phrases like "think hard" or "think carefully" start faster with no measurable drop in quality when cut. The practical issue facing enterprise budgets is direct: organizations are paying for surplus compute cycles while artificial coordination overhead delays real-world execution.
Adaptive Reasoning Replaces Legacy Directives
Topping the list of practices to abandon is Anthropic's advice to cut phrases like "think hard" or "think carefully" from prompts. The company explains that the model always reasons on its own and scales its reasoning depth to the task and the selected effort level. In tests conducted by Anthropic, eliminating these phrases allowed responses to start faster with no measurable drop in quality.
The model always reasons on its own and scales its reasoning depth to the task and the selected effort level.
That dynamic scaling removes the rationale for historical prompt hacks. Back when adaptive reasoning did not exist, users could sometimes coax more performance out of a model or steer the router this way. Today, those legacy strings simply inflate initial latency without generating any measurable quality dividend.
How does this dynamic alter configuration costs across routine corporate workloads?
Anthropic recommends starting at the "Medium" effort level, the default for Opus 5.5. In internal tests, Opus 5.5 on Medium matched the performance of its predecessor Opus 5 on "High." Higher settings cost more and don't always produce better results, meaning teams that default to maximum settings pay extra for negligible gains. Users can change the setting in the model menu next to the send button.
Model Economics and Execution Chains
The financial equation also challenges previous architectural compromises between model tiers. Teams that previously used the cheaper Sonnet and reserved Opus for complex projects should rethink that approach, according to Anthropic. The company argues that a more capable model can finish a task in fewer steps, which can partly offset its higher per-token price. As Anthropic noted, anyone who regularly cranks Sonnet up to "Extra High" or "Max" effort ends up paying about as much as with Opus 5.5.
This pricing reality builds on previous generational cost shifts. On 2026-08-10, Anthropic Claude Opus 5 halved the cost of frontier AI reasoning, delivering intelligence nearing the flagship Claude Fable 5 while squatting on the price point of its predecessor, Opus 4.8. In benchmarks published by Anthropic, Opus 5 more than doubles the performance of Opus 4.8 on Frontier-Bench v0.1—the gold standard for software engineering—while drastically lowering the cost per completed task. With Opus 5.5, operational efficiency depends heavily on avoiding multi-turn friction.
Anthropic advises describing the desired result in a single message rather than giving instructions one at a time, covering the goal, the output format, and when the model should check back with questions. Opus 5.5 is better than its predecessor at seeing multi-part tasks through, according to the company.
Some engineering teams may argue that legacy prompting habits provide fine-grained control that automated reasoning depth cannot reliably replicate across complex enterprise domains. However, Anthropic's data shows that manual workarounds often degrade efficiency. Workarounds such as manually transcribing chart values are no longer needed, as Anthropic stated. Even at the lowest effort level, it read values from dense charts more accurately than Opus 5 at the highest, according to the firm. Furthermore, users who need a finished document or spreadsheet should ask for it directly instead of requesting a draft first; as Anthropic explained, Opus 5.5 often goes from request to finished product in a single pass.
Open enterprise prompt templates this week, strip out instructional fillers like "think hard" or "think carefully", set default execution to Medium in the menu next to the send button, and configure complex tasks into a single comprehensive prompt accompanied by direct file attachments.
What came before
Where engineers once squeezed extra accuracy out of models with verbose steering prompts, Claude Opus 5.5 turns those legacy habits into pure latency and wasted token budgets. The system scales reasoning depth on its own, yet teams continue to pay for idle coordination overhead by pinning effort settings to the ceiling and clinging to obsolete trigger phrases. That friction eats directly into margins: maxing out Sonnet now rivals the total cost of running Opus 5.5, even as its lineage held the pricing of Opus 4.8 while cutting down the steps required to solve a task. Companies foot the bill for frontier autonomous reasoning while subsidizing prompts designed to micromanage it.
Our stories: Anthropic Claude Opus 5 Halves Cost of Frontier AI Reasoning (August 10).