Composer 2.5 has been released on the Cursor platform, offering an upgrade over Composer 2 in intelligence, reliability, and the ability to handle complex and long-running tasks. Aimed primarily at developers and professionals who require advanced AI coding assistance, Composer 2.5 is publicly available to all Cursor users. The model introduces architectural and training improvements, including targeted reinforcement learning with localized textual feedback, enabling more precise model corrections during long task rollouts.
Composer 2.5 is exceptionally intelligent and up to 10x more efficient than similarly capable models. pic.twitter.com/ie9X2VPjTU
— Cursor (@cursor_ai) May 18, 2026
Pricing for the standard version is $0.50 per million input tokens and $2.50 per million output tokens, while a faster variant, set as the default, is available at $3.00 per million input tokens and $15.00 per million output tokens.
Composer 2.5 is based on the same open-source checkpoint as its predecessor but incorporates a 25x increase in synthetic task training and improved behavioral calibration. These technical refinements address real-world requirements, such as communication style and coding consistency, making the model better at following nuanced instructions.
Cursor, in collaboration with SpaceXAI, is also developing a larger model using significantly more compute resources, hinting at future capabilities. Early feedback from users and internal benchmarks highlight Composer 2.5’s stronger performance in sustained coding tasks and more accurate handling of tool calls compared to previous versions and similar products on the market.