AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Anthropic says outdated prompting habits are slowing down Claude Opus 5.5

Collected Nov 14, 2026

Anthropic has made Claude Opus 5.5 the default model across its paid tiers, now including Pro, Max, Ultra and Team plans, and alongside that switch it published guidance telling users to abandon prompting habits carried over from earlier model generations. The company's central claim is that many users are actively holding the new model back with instructions that suited older releases.

The most prominent item on Anthropic's list is removing phrases such as "think hard" or "think carefully" from prompts. According to the company, Opus 5.5 always reasons on its own and scales reasoning depth to the task and to the effort level the user has selected. In its own tests, responses requested without those phrases started faster, with no measurable loss of quality. Anthropic traces the phrases to the early period of reasoning models, roughly a year and a half ago, when they could sometimes coax more performance out of a model or steer a router toward deeper processing. That workaround no longer applies, the company says.

Anthropic also recommends beginning at the Medium effort level, which is the default for Opus 5.5. Internal tests reportedly found that Opus 5.5 running at Medium matched the performance of its predecessor, Opus 5, at High. The company's advice is to move to High only when Medium proves insufficient, noting that higher settings cost more and do not always deliver better output. The setting sits in the model menu next to the send button. This reflects a broader pattern in frontier model releases: as reasoning becomes adaptive, the marginal gain from manual effort escalation shrinks, and paying for longer deliberation can be wasteful when the model would have reached the same answer more cheaply.

On economics, Anthropic argues that teams who historically reserved Opus for hard projects while running the cheaper Sonnet model elsewhere should reconsider. Its reasoning is that a more capable model can complete a task in fewer steps, which can partly offset a higher per-token price. Teams that routinely push Sonnet to Extra High or Max effort, the company says, end up spending about what they would have spent on Opus 5.5. In practice this is the classic cost-versus-steps trade-off: token prices are easy to compare, but total task cost depends on how many tokens and retries a given model needs, so a cheaper per-token rate can still lose over a full workflow. Anthropic's guidance leans on that distinction without offering specific pricing figures in this material.

A second cluster of recommendations concerns how work is framed. Anthropic suggests describing the desired end result in a single message rather than issuing instructions one at a time, covering the goal, the output format, and when the model should come back with questions. The company says Opus 5.5 is better than its predecessor at seeing multi-part tasks through and can increasingly handle the back-and-forth users used to conduct in chat, in effect prompting itself from one initial input. For developers building on the API, that implies prompt templates designed around tight multi-turn orchestration may need revision toward richer up-front goal statements.

On multimodal input, Anthropic recommends attaching diagrams and tables as images or files instead of transcribing them by hand, saying the workarounds older models required are no longer necessary. Image reading improved significantly in Opus 5.5, according to the company: even at the lowest effort level it read values from dense charts more accurately than Opus 5 at its highest. Users who want a finished document or spreadsheet should request it directly rather than asking for a draft first, since the model often goes from request to finished product in a single pass.

For design work, Anthropic advises naming the styles you want to avoid specifically, such as a cream-colored background or pill-shaped buttons, rather than issuing a vague ban on "generic design." General instructions in that vein, the company argues, merely swap one default style for another. Finally, connecting Claude to a team's own files and systems through connectors is said to produce significantly more tailored results, because the model can pull the information it needs itself rather than having it pasted into the conversation.

Why it matters: Developers and product teams running agents or batch jobs on Opus 5.5 are the people most affected, since prompt templates that inject "think harder" instructions or escalate effort levels may now add latency and cost without added quality, based on Anthropic's own tests. A likely inference is that evaluation suites built under older models should be re-run against Opus 5.5 at Medium effort before budget assumptions are locked in. Teams should also revisit routing logic that reserves Opus for hard tasks, since the cost argument now depends on task length and step count rather than token price alone.

Read at The Decoder

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Anthropic has made Claude Opus 5.5 the default model on all paid plans and published a guide that does away with old prompting habits. The main takeaway is to drop "think hard" and start at Medium effort. The article Anthropic says outdated prompting habits are slowing down Claude Opus 5.5 appeared first on The Decoder .