Opus 5.5 becomes Claude’s default on Pro, Max, Ultra, Team

Anthropic has made Claude Opus 5.5 the default model on Pro, Max, Ultra and Team plans. A Claude guide published November 13 recommends keeping it selected for research, analysis and writing, and for producing memos, reports, spreadsheets and decks. Teams that use Sonnet day to day and keep Opus for heavy analysis are told to flip that habit. Anthropic’s argument is that a more capable model finishes a task with less work, which offsets much of its higher per-token price. On those plans, though, higher-priced models still use up limits faster.

Billed by the token, Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, against $2 and $10 for Sonnet 5.5. The guide gives a clear point at which to switch. When Sonnet 5.5 runs at higher effort settings, its cost per task gets close to Opus 5.5, so anyone pushing Sonnet to Extra high or Max effort is told to move to Opus instead.

The tests behind the switch

The one per-task cost figure comes from an internal test, and it compares Opus 5.5 with Opus 5. Opus 5.5 analyzed a proposed merger between two fictional HR software companies, built a financial model in Excel and turned it into an executive presentation on whether the deal made sense at its price. It took 63 minutes, against 93 for Opus 5 on the same task, at half the cost. Anthropic says the financial model was also more thorough and the presentation easier to read.

A second internal test targeted invented numbers. Opus 5.5 wrote reports on a company’s quarterly results using only a saved snapshot of the web, in which the earnings release was hard to find. An automated grader checked every figure and quote against the snapshot, and a single invented one failed the whole report. Across effort settings, 16 of 18 reports passed. That still leaves two reports with at least one invented figure or quote.

On charts, Opus 5.5 at its lowest effort setting read values off dense charts more accurately than Opus 5 did at its highest. Frank Corrao, who leads central equity quant research engineering at Walleye Capital, said that at higher effort settings Opus 5.5 “detected that the minute indexing in our own instructions was off by one and corrected for it, noting that this would cost it points with the grader.” He added that no model his team had tested had caught and acted on that before.

Where Sonnet, Haiku and Fable still fit

The guide keeps a role for each of the other three models:

• Haiku 5.5, at $0.10/$0.50, is for quick lookups and high-volume work such as summaries and sorting data. Prompts over 100k tokens cost $0.50/$2.50. It is the first Haiku with effort controls, so you can ask it for a more careful pass before moving up. Longer analysis goes to Opus 5.5.
• Sonnet 5.5, at $2/$10, is for well-scoped tasks with a clear brief, like turning finished notes into a polished document or building a deck from a slide template.
• Fable 5.2, at $8/$40, is for the most demanding reasoning and long-running work, such as deep multi-step research or coding handed to a large team of subagents that run for hours.

On Pro plans, and on standard seats in Team and seat-based Enterprise plans, Fable 5.2 doesn’t count against plan limits outside limited-time launch offers. It runs on usage credits instead. That changes the meaning of the step Anthropic’s developer guide suggests for hard tasks already running Opus 5.5 at max. In that guide, on FrontierCode 2.0, Fable 5.2 at high effort scored 45.3% against Opus 5.5’s 45.5%, for roughly half the cost per task. On those seats, the same move takes the work off plan limits and onto paid credits.

The rule for moving between models applies to all of them. If Claude does worse than expected, raise the effort first. If it had the context, clearly tried and still got it wrong, move up a model.

Medium effort, and no “think hard”

Medium is the default effort for Opus 5.5. On Anthropic’s coding and knowledge-work evaluations, Opus 5.5 at Medium matched or beat Opus 5 at High. Higher settings think longer and cost more without always producing better work, so the guide says to move to High only when Medium stalls. The setting sits under the model name next to the send button, under Effort. Zimu Li, Member of Technical Staff at Factory, said: “Claude Opus 5.5 is the first model we’d default to at medium effort. In our testing it matched Opus 5 on high effort, while using 20 to 25% fewer output tokens.”

Thinking is always on for Opus 5.5, and the model decides how much to think from the request and the effort level. In Anthropic’s testing, deleting lines like “think hard” and “think carefully” made replies start sooner, with no clear drop in quality. The guide also asks for a few other changes:

• Put the goal, the finished output and the point where Claude should stop and ask in the first message, rather than giving instructions one at a time.
• Attach the chart, screenshot or diagram instead of typing its numbers into the prompt.
• For design work, name the patterns to avoid, such as a cream background or pill-shaped buttons. A general “avoid a generic look” tends to swap one default style for another.
• Connect Claude to your files and systems. On Team and Enterprise plans, an owner has to turn a connector on before members can use it.

Anthropic’s own demo used this kind of prompt. An analyst at a hypothetical retailer asked Opus 5.5, at Medium, to test leadership’s four theories about slowing growth against a folder of sales data, without saying which file to check for which theory. Claude found that none of the four held up, though one couldn’t be fully ruled out. It then pointed to a cause nobody had raised: new stores opened too close to existing ones were drawing sales away from them, and it backed that with a regression against the distance between stores.

The guide closes with a trial anyone can run. Take a task you’d normally split across several requests, such as a research memo, a variance analysis or a deck built from an existing doc. Give Opus 5.5 the files and a description of the finished result in one message, leave it at Medium effort, and compare the result with what other models give you.