Muse Wonder from Meta is now in Cursor's model picker
Cursor says it scores 46.1% on CursorBench, the highest of any Muse model. It costs $4 in and $20 out per million tokens, the same input and output rates as Opus 5.5.
Muse Wonder is now available in Cursor. Cursor announced it at 6:56 PM ET on Oct 15, saying it "scores 46.1% on CursorBench at default effort, the highest of any Muse model," and prices it at $4 per million input tokens and $20 per million output tokens. Those input and output rates match Claude Opus 5.5 on Cursor's price table. Cursor's docs say it is a third-party model, so it "isn't part of the usage included with Cursor plans."
What changed#
The model page calls it Meta's flagship and the successor to Muse Spark 1.3 in Cursor, built for agentic coding: long tool-use chains, multi-step debugging, and large-repository work.
Cursor's docs list these details:
- Model ID
muse-wonder-1.0, provider Meta. - 300k token context window by default, or 1M in Max Mode at the same per-token rates.
- Six effort levels: minimal, low, medium, high (the default), extra high, and max.
- Image input, including screenshots returned by tools.
- Access to all agent tools when used in Cursor.
- Pricing: $4 input, $0.20 cache read, $20 output per million tokens. No cache-write charge and no long-context multiplier.
- On The X Subscription it bills from prepaid credit. On legacy Cursor plans it draws from the Other Models pool.
Meta's Muse Wonder blog says it is available through the Meta Model API and Muse Code. Tonight's news is that it is now in Cursor.
Why it matters#
Our read: if you use Muse Spark 1.3 or want a Meta flagship in Agent, Wonder is the top Muse on Cursor's own benchmark at default effort. At $4/$20 it matches Opus 5.5's input and output rates, so treat it as a paid third-party pick rather than a free upgrade inside your weekly pool.
How to use it#
- Choose Muse Wonder in the model picker. Its default effort is high.
- Teams and Enterprise admins who configured model access before Meta was available must enable the Meta provider before members see Wonder, per the model page.
- Per the docs, "drop to low or minimal for faster, cheaper responses on simpler tasks"; higher efforts "spend more reasoning tokens per step and score higher on CursorBench."
- Our suggestion: confirm prepaid credit (X Subscription) or Other Models pool headroom (legacy) before long runs.
Prompt (ours, untested):
Using Muse Wonder at high effort, fix the failing test in this repo, explain the root cause in three sentences, and list every file you changed.
Good to know#
- Wonder is not part of included Cursor usage. Per Cursor's plans table, Free and Plus get no third-party models; Premium and above pay API list price from prepaid credit.
- Cached input is $0.20 per million tokens. Max Mode above 300k uses the same per-token rates with no long-context surcharge.
- It is not a Grok Bot model choice. Cursor's docs say Grok Bot manages model selection, so there is no model picker there.
What to watch#
Whether independent CursorBench-style results hold the 46.1% lead over other Muse models, whether Wonder appears in Auto or Cloud Agents defaults, and whether Teams that already use Meta models need any further admin steps beyond enabling the provider.
