Claude Haiku 5.5 in Cursor: Pricing and Uses

Claude Haiku 5.5Anthropic's small, low-cost model in Cursor at $0.10 per million input tokens and $0.50 per million output for requests up to 100k input tokens, with 5x rates above that. Press Enter for the full definition. is the fastest model in Anthropic's Claude 5.5 family, and it is the successor to Claude 4.5 Haiku. Cursor's docs give it 48.4% on CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. at max effort, on the same cost-performance curve as Sonnet 5.5 and Opus 5.5, at a tenth of Haiku 4.5's per-token price for requests up to 100k input tokens. You add it to the model picker from Cursor Settings > Models.

Cursor describes Haiku as sitting on the same cost-performance curve as Sonnet 5.5 and Opus 5.5, while also saying those two score higher on CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. for harder tasks.

The model page lists four strengths and three limitations, so it is a more balanced description than a launch announcement. The strengths are low cost for high-volume work, effort levels, the option to turn thinking off, and context windows up to 1M tokens.

It is also the first Haiku with effort levels. Effort and the thinking switch are covered below, because together they account for the spread in its published benchmark scores.

Context is where the two numbers in this article can be confused. Haiku 5.5 supports context windows up to 1M tokens, while the price changes at 100k input tokens. The window tells you what the model can hold, and the 100k line tells you what you will pay per token to hold it, so a large window is available without being cheap.

Haiku 5.5 is a third-party model, so it is not 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 bills at the rates below 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. All prices are per million tokens.

The billing route is worth knowing before you add it to the picker. Because it sits outside the included usage, a Haiku run does not draw down your plan's own allowance. It is charged on its own meter, either your prepaid credit or 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, depending on which plan you are on.

Cursor's models and pricing page adds that third-party models are available on Premium and above at the model's API list price from the first request, paid from prepaid credit, and that Plus and Free do not include them. Top-ups add the credit, Auto Top-up refills it below a threshold you set, and a monthly additional usage limit caps paid additional usage each month.

Every rate in the right-hand column is exactly 5x the standard one once a request crosses 100k input tokens.

The models and pricing page lists the same Haiku 5.5 rates and adds that on legacy request-based plans it requires Max Mode. It also describes Haiku 5.5 as 90% cheaper than Claude 4.5 Haiku on input and output for requests up to 100k input tokens, and Claude 4.5 Haiku's listed $1 input and $5 output are ten times $0.10 and $0.50.

Cursor's docs call this out for long agent sessions: sessions with large contexts cost more per token once they cross that line.

Haiku 5.5 is the first Haiku to offer effort levels. With thinking on, 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 climbs from 30.9% at low effort to 48.4% at max, and Cursor offers a thinking variant for deeper reasoning. Thinking can also be turned off at low, medium and high effort for faster, cheaper responses.

The table gives two different dials, and they behave differently. Effort is the wide one: the same model spans 30.9% to 48.4% with thinking on. Thinking is the sharp one, because turning it off drops the score at every effort level where the option exists.

So the speed you buy by turning thinking off is paid for in quality, and the docs say so plainly: the off scores sit below the on scores at the same effort. Cursor lists the lower quality with thinking off as a limitation.

One thing the source does not give is latency or cost numbers for each effort setting, only benchmark scores. Any speed gain from lower effort is therefore not a figure Cursor has published.

Cursor lists low cost for high-volume work as a strength of Haiku 5.5, and lists the higher CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. scores of Sonnet 5.5 and Opus 5.5 on harder tasks as a limitation.

The cost of choosing Haiku shows up on the hard tasks. Cursor says Sonnet 5.5 and Opus 5.5 score higher on CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. there, so the saving is real only if the cheaper model's result is good enough for the job. That is a call about your own tasks, and I can't make it from the benchmark alone.

The tool story is the simple part. Haiku 5.5 has access to all agent tools when used with Cursor, so choosing it does not cost you capabilities the larger models have, and the decision stays about quality, speed and price. Our guide to choosing a Cursor model is the place to compare it with the rest of the lineup.