Claude Sonnet 5.5 in Cursor: Pricing and Uses

Claude Sonnet 5.5Anthropic's medium-tier model in Cursor at $2 per million input tokens and $10 per million output, with cache reads at $0.10; it keeps Claude Sonnet 5's input and output rates and succeeds it. Press Enter for the full definition. is Anthropic's latest medium-tier model and the successor to Sonnet 5. Cursor's documentation puts its CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. score at 55.5% at max effort, second only to Opus 5.5, and says it keeps Sonnet 5's input and output rates while halving the cache-read rate. You add it to the model picker from Cursor Settings > Models.

The headline claim is frontier-level CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. results at Sonnet pricing, and Cursor says Sonnet 5.5 sits on the same cost-performance frontier as Opus 5.5. I read that as a statement about value per dollar rather than about the ceiling, since the same page sends the hardest tasks to Opus 5.5.

Sonnet 5.5 has access to all agent tools when used with Cursor. The model page links to Cursor's agent overview and its tool-calling fundamentals for how the tools work.

If Sonnet 5 is already your pick, the rate card changes in one place only. Input and output stay where they were, and the cache-read line is the single number that moved.

Effort is the main dial Cursor's documentation attaches to this model. The CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. score climbs from 35.8% at low effort to 55.5% at max, a gap of 19.7 points on the same model, and the page frames this as a way to trade tokens for quality on harder tasks.

The two numbers in that table are easy to read together and, as it turns out, they shouldn't be. The 19.7 points runs from low to max, while the 6x runs from the default high to max, and since the doc doesn't publish the score at high you can't divide one by the other.

Six times is a big multiple, and it comes from CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. tasks, which may not look much like yours. The doc doesn't say how it carries over to other work, so it is best read as a published data point for that benchmark.

Higher effort levels use many more tokens, and that is the mechanism behind both the score gain and the bill. Cursor presents it as a choice you make per task: spend more tokens when the task is hard enough to use them.

Sonnet 5.5 bills at $2 per million input tokens and $10 per million output tokens, the same as Sonnet 5. Prompt-cache reads are $0.10 per million tokens, half of Sonnet 5's $0.20. All prices on the page are per million tokens.

Because Sonnet 5.5 is a third-party model, it isn't part of the usage included with Cursor plans. On The X SubscriptionThe subscription Cursor is sold through: one plan for Grok, Cursor and X, with Free, Plus, Premium, Super and Ultra tiers, per-seat Teams, a weekly included pool for Cursor models shared with Grok, and third-party models billed at API list price from prepaid credit. Press Enter for the full definition. it is billed at these rates from your prepaid credit, and on legacy Cursor plans it draws from the Other ModelsOn legacy Cursor plans, the monthly pool for supported third-party models from providers such as Anthropic, Google and OpenAI, separate from Cursor Models. On The X Subscription third-party models are billed from prepaid credit instead. Press Enter for the full definition. pool.

Cursor's models and pricing page adds that third-party models are available on Premium and above, and that Plus and Free do not include them. That page also lists a cache-write rate for Sonnet 5.5 of $2.50 per million tokens, alongside the $2 input, $0.10 cache-read and $10 output rates, and says Sonnet 5.5 requires Max Mode on legacy request-based plans.

Halving the cache-read rate should matter most where a long session keeps re-reading the same context, though the doc doesn't quantify the saving and I'd guess it varies a lot by workload. At $0.10 against $2 for fresh input, a cache read costs one twentieth of the input rate. That ratio is my arithmetic from the table, not a figure Cursor quotes.

The doc is explicit that all Sonnet 5.5 prompts bill at the base per-token rates, including when context goes above 200k, with no separate long-context multiplier. Tokens still cost money, though. At the $2 input rate a 500k-token prompt is $1 of input before any cache reads or output, which is plain multiplication and not a number from the page.

US-only endpoints are priced 10% higher, at $2.20 per million input and $11 per million output, while global endpoints use the base rates. Cursor points to its Privacy and Data Governance page for data residency details.

Cursor says Opus 5.5 remains the stronger choice for peak quality on the hardest tasks, and Sonnet 5.5's own 55.5% sits second to it. The page doesn't repeat Opus 5.5's score or rates, so any comparison from this page alone is only as precise as that one sentence.

Cursor lists two limitations for Sonnet 5.5, and they pull in different directions. The Opus 5.5 limitation argues for paying more per token on the hardest work. The effort limitation argues for paying more per task on the model you already have.

What the doc stops short of is a rule for choosing between them. It doesn't say whether max-effort Sonnet 5.5 or default-effort Opus 5.5 is the better buy on a given task.

Sonnet 5.5 supports thinking mode, and a thinking variant is available for deeper reasoning. Whether that variant behaves like higher effort on the same model or is a separate entry in the picker, the page doesn't say.