Meta Muse Wonder Lands on Cursor: 46.1% on CursorBench

Meta Muse Wonder Lands on Cursor: 46.1% on CursorBench

Meta Superintelligence Labs' model Muse Wonder has officially launched on the Cursor platform, giving developers a new coding assistance option. According to official information, the model scored 46.1% on Cursor's own CursorBench benchmark at default effort — the highest score among all Muse series models.

Performance: 46.1% on CursorBench

CursorBench is Cursor's internal benchmark for evaluating model coding capabilities. Although its exact composition is not fully disclosed, a score of 46.1% places it at the top of the Muse family. For developers watching AI coding tools, this result suggests that Muse Wonder may deliver stronger overall performance on tasks such as code generation, completion, and debugging. However, given the lack of cross-vendor comparative data, the practical significance of this score still needs to be verified through more independent testing.

CursorBench is an evaluation benchmark built by Cursor specifically to measure how large language models perform in real-world coding scenarios. It typically includes tasks such as code completion, function generation, bug fixing, and multi-file editing, scored via automated tests or comparison against reference implementations. Because it is an internal platform benchmark, the problem composition and scoring details are not fully public, so score comparisons between different models should be interpreted with caution.

"Default effort" refers to the default compute budget or chain-of-thought length the model uses during inference — not the highest setting. Vendors typically offer adjustable reasoning intensity: higher effort means the model generates more intermediate reasoning steps, potentially improving accuracy but increasing latency and token consumption. So 46.1% reflects performance under standard settings, not the model's capability ceiling.

Pricing: $4 per million input tokens, $20 per million output tokens

Muse Wonder uses token-based billing: input is priced at $4 per million tokens and output at $20 per million tokens. This pricing sits in the mid-to-high range for current AI coding models, with output costs significantly higher than input — consistent with the usual billing logic for generative models. For developers who use Cursor heavily, cost will be an important factor in model selection.

In token-based billing, tokens are the basic units of text that the model processes, roughly equivalent to part of an English word or a single Chinese character. Input tokens refer to the prompts, code context, and file content that users send to the model; output tokens are the responses, code, and suggestions the model generates. Output prices are usually several times input prices because generation requires token-by-token autoregressive computation, which is more compute-intensive.

For example, with input at $4 per million tokens and output at $20 per million tokens, a request containing 10,000 input tokens and 2,000 output tokens would cost approximately $0.04 + $0.04 = $0.08. For developers who call the model frequently throughout the day for code completion, cumulative costs can rise quickly — making context-length management and model-switching strategies critical for controlling expenses.

What It Means for Developers: More Model Choices and Cost Trade-offs

The addition of Muse Wonder gives Cursor users more model options. Developers can switch between Muse Wonder and other available models based on task complexity, budget, and performance needs. Its high CursorBench score may attract users focused on code generation quality, while the pricing reminds teams to assess how usage volume affects costs.

Market Positioning and Competitive Landscape

Meta Superintelligence Labs' launch of Muse Wonder on Cursor signals its intent to enter the AI coding tools market. As a popular AI code editor, Cursor has become a platform where model vendors showcase their capabilities. Muse Wonder enters this highly competitive space with the highest score in its family and clear pricing — whether it wins over developers will depend on real-world experience and cost-effectiveness.

Currently, details about Muse Wonder's architecture, training data, and broader benchmark results have not been disclosed. Developers may want to watch for real-world trial feedback on the Cursor platform.

Meta Superintelligence Labs is an organization established by Meta to advance frontier AI research, focusing on general intelligence and large-scale model development. The Muse series is its model family for developer and productivity scenarios, and Wonder is the member of that family aimed at coding assistance. Integrating with third-party developer tools like Cursor means model vendors can bypass building their own apps to reach professional developers directly, while also using real platform usage data to iterate on their models.

The AI code editor space already includes multiple model providers, and platforms typically integrate several models at once so users can choose based on task and budget. This multi-model landscape reduces lock-in to any single model and makes pricing, latency, and benchmark scores like CursorBench key dimensions of vendor competition.

One-line Summary

Meta's Muse Wonder lands on Cursor with a 46.1% CursorBench score and mid-to-high pricing.

Paragraph Summary

Meta Superintelligence Labs' coding model Muse Wonder has officially launched on the Cursor platform. It scored 46.1% on Cursor's internal CursorBench at default effort — the highest in the Muse series — but since benchmark details are undisclosed and cross-vendor comparisons are lacking, the score's real significance still needs independent verification. Pricing is $4 per million input tokens and $20 per million output tokens, placing it in the mid-to-high range for AI coding models, with output costs significantly higher than input. Muse Wonder gives Cursor users more model choices, requiring developers to balance code quality, budget, and usage volume. The move also signals Meta's intent to enter the AI coding tools market; whether it wins over developers will depend on real-world experience and cost-effectiveness.