はてなブックマーク

These quality improvements come from our first continued pretraining run, which provides a far stronger base to scale our reinforcement learning. From this base, we train on long-horizon coding tasks through reinforcement learning. Composer 2 is able to solve challenging tasks requiring hundreds of actions. #Try Composer 2 Composer 2 is priced at $0.50/M input and $2.50/M output tokens. There is a

We train Composer for long-horizon tasks through a reinforcement learning process called self-summarization. By making self-summarization part of Composer's training, we can get training signal from trajectories much longer than the model's max context window. This translates into Composer being able to learn to work on challenging coding tasks requiring hundreds of actions. #The limits of compact

Developers are asking coding agents to take on longer, more complex tasks that span multiple files, tools, and steps. As these requests grow in scope, the evals that measure agent performance need to evolve with them. At Cursor, we use a hybrid online-offline eval process to keep our understanding of model quality aligned with what developers actually do. The offline part uses CursorBench, our int

Feb 18, 2026 by Ani Betts, Yash Gaitonde & Alex Haugland in research Coding agents are getting much better at running terminal commands to explore the environment and make changes. Users who auto-approve these commands unlock significantly more powerful agents, but at the cost of increased risk. A mistaken agent can delete databases, ship broken code, or leak secrets. Requiring human approval for

Semantic search is one of the biggest drivers of agent performance. In our recent evaluation, it improved response accuracy by 12.5% on average, produced code changes that were more likely to be retained in codebases, and raised overall request satisfaction. To power semantic search, Cursor builds a searchable index of your codebase when you open a project. For small projects, this happens almost

As coding agents became more capable, we found ourselves spending more time on review. To solve this, we built Bugbot, a code review agent that analyzes pull requests for logic bugs, performance issues, and security vulnerabilities before they reach production. By last summer, it was working so well that we decided to release it to users. The process of building Bugbot began with qualitative asses

We've been experimenting with running coding agents autonomously for weeks. Our goal is to understand how far we can push the frontier of agentic coding for projects that typically take human teams months to complete. This post describes what we've learned from running hundreds of concurrent agents on a single project, coordinating their work, and watching them write over a million lines of code a

Coding agents are changing how software gets built. Models can now run for hours, complete ambitious multi-file refactors, and iterate until tests pass. But getting the most out of agents requires understanding how they work and developing new patterns. This guide covers techniques for working with Cursor's agent. Whether you're new to agentic coding or looking to learn how our team uses Cursor, w

Coding agents are quickly changing how software is built. Their rapid improvement comes from both improved agentic models and better context engineering to steer them. Cursor's agent harness, the instructions and tools we provide the model, is optimized individually for every new frontier model we support. However, there are context engineering improvements we can make, such as how we gather conte

We're excited to release a visual editor for the Cursor Browser. It brings together your web app, codebase, and powerful visual editing tools, all in the same window. You can drag elements around, inspect components and props directly, and describe changes while pointing and clicking. Now, interfaces in your product are more immediate and intuitive, closing the gap between design and code. #Rearra

Composer is our new agent model designed for software engineering intelligence and speed. On our benchmarks, the model achieves frontier coding results with generation speed four times faster than similar models. We achieve these results by training the model to complete real-world software engineering challenges in large codebases. During training, Composer is given access to a set of production

Today, we’re releasing two big updates that make Cursor the best place to work with agents: our first coding model, Composer, and a new interface for working with many agents in parallel. #Introducing Composer Composer is a frontier model that is 4x faster than similarly intelligent models. The model is built for low-latency agentic coding in Cursor, completing most turns in under 30 seconds. Earl