- Claude Opus 5.5 is the default model on Pro, Max, Ultra and Team plans. It costs $4/$20 per million tokens, twice as much as Sonnet 5.5.
- Anthropic advises starting at Medium effort and deleting “think hard” from prompts: the company says replies start sooner with no clear quality drop.
On November 12 Anthropic published a guide urging people to use Claude Opus 5.5 as their everyday model rather than the cheaper Sonnet. Opus 5.5 is already the default on Pro, Max, Ultra and Team plans. The main argument is counter-intuitive: a stronger model finishes a task in fewer steps, so a higher price per piece of text does not have to mean a higher bill. The company also lists prompting habits that now only slow the work down.
The price list side by side
Anthropic quotes prices per million tokens, input and output. A token is a piece of text that the price is measured in.
- Haiku 5.5: $0.10 and $0.50
- Sonnet 5.5: $2 and $10
- Opus 5.5: $4 and $20
- Fable 5.2: $8 and $40
So Opus costs exactly twice as much as Sonnet for every token. On subscription plans you pay differently: a pricier model simply uses up your limits faster.
The maths: when Opus comes out cheaper
Because the price is double, Opus has to use fewer than half of Sonnet’s output tokens to come out cheaper. Here is my own simple example using the list prices: if Sonnet writes 50,000 output tokens on a task, the output costs you $0.50. Opus would charge $1 for the same 50,000, and only breaks even with Sonnet at 25,000.
Is that realistic? Anthropic shows its own tests. On an analysis of a fictional merger, with a financial model in Excel and a board presentation, Opus 5.5 finished in 63 minutes against 93 for the previous Opus 5, and according to the company at half the cost. Factory, quoted in the guide, says Opus 5.5 at Medium matched Opus 5 at High while using 20 to 25% fewer output tokens. But that compares it with the old Opus, not with Sonnet 5.5. Anthropic does not say how many tokens Sonnet uses on the same tasks, so the question “will it be cheaper for me” can only be answered by your own numbers. One hint from the guide: if you often push Sonnet to Extra high or Max effort, the cost per task gets close to Opus anyway.
Habits to drop
The effort level decides how long the model thinks. Anthropic advises starting at Medium, which is the default, and moving to High only when Medium stalls. According to the company, higher settings cost more and do not always produce better work.
- Delete “think hard” and “think carefully”. Opus 5.5 always thinks, and in Anthropic’s testing replies started sooner without those lines, with no clear drop in quality.
- In your first message, describe what the finished result looks like and when the model should stop and ask, instead of feeding instructions one at a time.
- Attach the chart or screenshot instead of retyping its numbers.
- Name the specific styles you do not want, such as a cream background, rather than a vague “avoid a generic look”.
According to Anthropic, the biggest gain comes from a connector, a link to your files and systems that lets the model fetch what a task needs by itself. On Team and Enterprise plans an owner has to switch a connector on first.
How to check it on your own work
Take one typical task that you now split across several messages and run it once in Sonnet and once in Opus at Medium, with the same description of the finished result. Compare the token count on the bill and the time you spend fixing the output. I write about this in more depth in my article on task cost in tokens, and I track other models’ prices in the Model pricing hub. In my view the “Opus every day” advice makes sense for longer document-heavy tasks, while for short, repetitive questions Sonnet or Haiku will still be cheaper.

