Muse Wonder in Cursor: Pricing and Uses

Muse Wonder is Meta's flagship model in Cursor and the successor to Muse Spark 1.3. Cursor's documentation describes it as built for coding work that runs long: long tool-use chains, multi-step debugging and large-repository work.

Cursor's page doesn't publish a benchmark number for Muse Wonder, so the description is a statement of intent and not a measured result.

Muse Wonder has access to all agent tools when used with Cursor. The model page points to Cursor's agent overview and its tool-calling fundamentals for the details, so this is the place to read about what those tools actually are.

It also takes image input, including screenshots returned by tools. The page states the capability and does not describe a use case.

Muse Wonder gives you two dials: how hard the model thinks, and how much of your repository fits in one request. Effort comes in six levels, and the context window has a standard size and a larger one.

The mechanism behind effort is spelled out in one clause: higher efforts spend more reasoning tokens per step. That is why the same sentence ties better CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition. scores to higher effort and also tells you to drop to low or minimal when the task is simple. You pay for the thinking, and a rename of one variable doesn't need it.

The doc doesn't publish 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 at each level, and it doesn't give a cost multiple between levels the way some other model pages do. So the direction is clear and the size of the trade is not.

On context, the 1M window needs Max Mode, and the rates don't change when you switch it on. Larger windows still cost more in practice, because a bigger prompt is more tokens at the same price per token. The next section works through that.

Muse Wonder lists at $4 per million input tokens and $20 per million output tokens, with cached input at $0.20 per million. All prices on the page are per million tokens.

Muse Wonder is a third-party model, so 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 like Claude, GPT and Gemini 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 prepaid credit, Auto Top-up refills it below a threshold you set, and a monthly additional usage limit caps paid additional usage each month.

Compared with its predecessor, the input rate is about three times higher and the output rate nearly five times, which is my arithmetic from the table and not a figure Cursor states. Muse Spark 1.3 is marked hidden by default on the pricing page.

The pricing table also notes that Muse Spark 1.3 requires Max Mode on legacy request-based plans. For Muse Wonder it notes up to 1M tokens in Max Mode at the same per-token rates, with cached input billed at $0.20 per million and no separate cache-write charge.

Prompts bill at the base per-token rates, including in Max Mode above 300k tokens. Cursor says there is no long-context multiplier and no cache-write charge. A 500k-token prompt at the $4 input rate is $2 of input before any cache reads or output, which is plain multiplication and not a number from the page.

A cache read at $0.20 is one twentieth of the input rate, again by my arithmetic. That gap should matter in long sessions that keep re-reading the same context, though the doc doesn't quantify the saving and it probably varies a lot by workload.

For Teams and Enterprise, the model page carries a notice at the top about the Meta provider, and it is easy to miss if your admin set up model access a while ago.

The notice is about timing: the configuration predates Meta as a provider, so an admin has to enable the provider. Cursor links to the model access control section of its enterprise documentation for the setting itself.

On data, Muse Wonder runs under Cursor's standard provider terms. Cursor points to Meta's zero data retention policy and to its own data use overview. The page doesn't restate what either says.

The page lists no limitations section for Muse Wonder, unlike some model pages. That is an absence in the documentation and not evidence that there are none.

The strengths the page does list are the six effort levels, the 300k token window with 1M in Max Mode at the same per-token rates, and image input including screenshots returned by tools.