Anthropic’s 12 November guide says Opus 5.5 is the default on Pro, Max, Ultra and Team, and says to delete “think hard”
Aleksandra Todorova’s article prices Opus 5.5 at $4 and $20 per million tokens, against $8 and $40 for Fable 5.2. In the company’s merger test, Opus 5.5 finished in 63 minutes and Opus 5 in 93, at half the cost. A Factory engineer says medium effort used 20 to 25 percent fewer output tokens.
Feature3 minEasy
Priya Shah3 min read
- Anthropic
- Claude
- Opus 5.5
- Prompting

Photo: AnthropicSource page
Anthropic says Claude Opus 5.5 is now the model already selected on its Pro, Max, Ultra and Team plans, and that people who had been saving Opus for the hard jobs should look again. Aleksandra Todorova’s guide, dated 12 November on Claude’s site, calls Opus 5.5 the daily driver for research, analysis, and for memos, reports, spreadsheets and decks. The reason it gives for switching off Sonnet for ordinary work is that a more capable model can finish a task by doing less work, which can offset much of a higher price per token. The price table on the page is $4 per million input tokens and $20 per million output tokens for Opus 5.5, against $2 and $10 for Sonnet 5.5, $8 and $40 for Fable 5.2, and $0.10 and $0.50 for Haiku 5.5 on prompts up to 100,000 tokens. Longer Haiku prompts, the footnote says, cost $0.50 and $2.50. On those four paid plans, higher-priced models use the allowance faster.
The habit the guide wants deleted is a pair of phrases. “Delete ‘think hard’ and ‘think carefully.’” Thinking is always on for Opus 5.5, it says, and the model decides how much from the request and the effort setting. In Anthropic’s testing, removing those lines made replies start sooner, with no clear drop in quality. Medium is the default. On the company’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 when Medium stalls. The control it names is the model name next to the send button.
Sixty-three minutes, and a chart it says you should not retype
Two of the tests are internal, and the page says so. In one, Opus 5.5 wrote reports on a company’s quarterly performance from a saved copy of the web in which the earnings release was hard to find. An automated grader checked every figure and quote against that snapshot, and a single invented figure or quote failed the report. Across effort settings, 16 of 18 reports passed. In another, the model analyzed a proposed merger of two fictional HR software companies, built a financial model in Excel, and turned it into an executive presentation. It finished in 63 minutes, against 93 for Opus 5 on the same task, at half the cost, with what Anthropic calls a more thorough model and an easier presentation to read. On vision, the guide says that even at its lowest effort setting Opus 5.5 read values off dense charts more accurately than Opus 5 did at its highest, and that transcribing a chart into the prompt may no longer be needed.
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.
— Zimu Li, member of technical staff at Factory, in Anthropic’s guide
Two other quotations are from customers, not from the lab’s own grader. John Ruelas, a staff software engineer at Ramp, said: “It writes like a good colleague, and follows our writing rules. A design spec came out usable with very minimal edits, and when it rewrote one of our prompts I preferred its version to my own.” Frank Corrao, who leads central equity quant research engineering at Walleye Capital, said that at higher effort settings the model “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. It was right, and no model we’ve tested had caught and acted on that before.” The guide’s other instructions are practical. Describe the finished result in the first message, including when the model should stop and ask. For design, name the styles to avoid, such as a cream background or pill-shaped buttons, because a general ban on a “generic look” tends to swap one default for another. Fable 5.2, it says, is for the most demanding reasoning and for long runs, and on Pro plans and standard Team seats it is not inside the plan limits outside limited-time launch offers, so it runs on usage credits.


