Cursor integrates Muse Wonder from Meta Superintelligence Labs, scores 46% on CursorBench

Summary

Meta Superintelligence Labs has launched Muse Wonder in the Cursor coding environment, marking it as their most capable model for personal and professional agent tasks. It achieved a score of 46.1% on CursorBench at default effort, the highest among Muse models, and is priced at $4 per million input tokens and $20 per million output tokens. The Muse models are designed to enhance memory retention and execution focus, which support long-running agent workflows, making them a significant addition for developers using Cursor and Meta's tools.

Analysis

Cursor: Cursor is an AI-powered coding agent and development environment that integrates advanced language models to assist developers in building software. It has incorporated Meta's Muse Wonder model, enabling users to leverage its capabilities directly within the platform for coding and related tasks. The integration highlights Cursor's role in making cutting-edge AI models available to builders. Muse Wonder: Muse Wonder is a frontier AI model series developed by Meta Superintelligence Labs, optimized for state-of-the-art performance on agentic, memory-intensive, and coding tasks. It introduces features like Active Memory and Active Focus to improve reliability in handling complex, multi-step workflows. The model is now available in Cursor, expanding access for developers working on ambitious software projects. Meta Superintelligence Labs: Meta Superintelligence Labs is Meta's dedicated AI research and development division focused on advancing toward personal superintelligence. It recently released the Muse Wonder series of frontier models designed for complex agentic tasks in personal and professional settings. The lab's latest models, including Muse Wonder, are now accessible via developer tools and platforms such as Cursor. Model Release: Meta Superintelligence Labs has positioned Muse Wonder as its most capable models to date for personal and professional agent tasks. Agent Capabilities: The models emphasize improved memory retention and focused execution to support long-running agent workflows. Platform Integration: Developers can now access Muse Wonder through the Cursor coding environment alongside Meta's own API and Muse Code tool.

Categories

techaiai_agentsmachine_learning

Related sources

View Original Tweet