The latest AI news and updates of Cursor — the AI-first code editor built for pair-programming with AI. Covering Cursor's latest product updates, launches, and company news from the past 90 days.
Cursor13h agoCursor has added DeepSeek V4.1 Flash to its model picker. The model occupies the CursorBench cost frontier, with pricing ranging from $0.60 to $0.98 per task. At these price points, no other model scores higher on CursorBench. The model is available for selection in Cursor Settings under the Models menu.
Cursor21h agoCursor released the Agent SDK, which defines agents as code using Markdown prompts and TypeScript tools. The SDK supports deployment on Cursor-managed or self-hosted infrastructure and includes built-in evals to measure and improve agent performance. It uses the same harness that powers Cursor's internal agents like Bugbot, Security Review, and Rollouts.
CursorOct 23Cursor agents now operate desktop applications beyond the browser, enabling autonomous interaction with tools like the iOS Simulator. The agent navigates interfaces, clicks, and types to verify features during development. This capability triggers automatically when a task requires it or through an explicit @-mention in the prompt.
CursorOct 16Cursor cloud agents now generate interactive canvases for dashboards and reports, which can be shared with teams via link. The new beta Environments API enables programmatic management of agent environments, including repository cloning and configuration. Additionally, cloud agent environments now support up to 500 repositories, with large multi-repo environments starting 5x faster.
CursorOct 16Cursor has added Meta Superintelligence Labs' Muse Wonder model to its AI code editor. The model scores 46.1% on CursorBench at default effort, the highest result among Muse models. It is available for coding tasks at a price of $4 per million input tokens and $20 per million output tokens.
CursorOct 13Cursor has added Google's Gemini 4 Argon to its AI code editor. The model scores 52.8% on CursorBench at default effort, performing on par with Opus 5.5. It is available for coding tasks at a price of $2 per million input tokens and $10 per million output tokens.
CursorOct 13Cursor’s Origin code hosting platform now supports over 20 integrations across CI, deployments, code review, security, monitoring, and project management. The Origin App Marketplace now accepts public submissions for developer tools. Listed apps require a 5 to 10-minute demo video and a detailed description for Cursor’s review process.
CursorOct 8Cursor has added Anthropic's Claude Haiku 5.5 to its model picker. The model costs $0.10 per million input tokens and $0.50 per million output tokens for requests up to 100k tokens, with higher rates for long-context sessions. Additionally, Cursor reduced Claude Sonnet 5.5 cache read pricing to $0.10 per million tokens.
CursorOct 7Cursor launched Remote Control, providing mobile access to local agents from the Cursor iOS app. Agents continue running on the computer, maintaining operation even if the phone loses signal. The feature is active by default for all accounts except Enterprise organizations, and includes a desktop setting to prevent the computer from sleeping during remote sessions.
CursorOct 5Cursor updated its SDK to allow steering agents mid-task with run.steer(), which moves subagents to the background to continue working. Background subagents now return results to the parent on the same run. The update also adds MCP annotations for custom tools, a replaceable system prompt rolling out account-by-account, and local TypeScript and Python support for background results.
CursorOct 2Cursor Rollouts now automates regression remediation by identifying the offending pull request and opening an issue upon detection. A single click initiates a cloud agent to apply a fix. Rollouts usage credits are included through October 3, alongside the existing monitoring plan and deployment verification capabilities.
CursorOct 2Cursor has added GLM 5.3 and GLM 5.3 Flash to its AI-first code editor. The update includes new CursorBench 4.0 results, where Cursor identifies GLM 5.3 Max as the best-scoring open-weight model. These models are now selectable for coding tasks directly within the editor.
CursorSep 29Cursor now builds charts and diagrams directly within the chat. The new /visualize command analyzes data and returns the resulting visualization inline. This feature is available now within the Agents Window, where it generates visual outputs for data analysis tasks.
CursorSep 28Cursor has added Anthropic's Claude Sonnet 5.5 to its AI code editor. The model performs on par with Opus in many tasks according to Cursor's internal testing. The update includes new CursorBench 4.0 results comparing the model's performance and cost against other available AI models.
CursorSep 23Cursor launched Rollouts, a bot that writes monitoring plans and verifies deployments to catch regressions before users see them. The update also improves Security Reviewer, which now finishes in 3.8 minutes on average, a 21% speed increase. Both tools are available today on Teams and Enterprise plans, with 10 days of Rollouts usage credits included.
CursorSep 23Cursor reduced token costs for its AI agents by 7% without impacting quality. The team achieved these savings by trimming system prompts, loading tools only when needed, improving cache reuse, and compressing file reads. These harness-level optimizations allow agents to complete more work while consuming fewer tokens per run.
CursorSep 22Cursor has added Claude Opus 5.5 to its AI code editor. The model ranks as the new top performer on CursorBench with a 57.8% (Max) score. It also reduces costs by 40% per task compared to the previous Opus 5 release.
CursorSep 11Cursor launched Projects in beta, a persistent thread where a coordinator agent manages subagents to handle complex tasks. Projects sync shared memory and artifacts across devices and support event-driven subscriptions, such as monitoring Slack or following PRs to fix CI issues. The system handles larger bodies of work, with primary users merging six times as many pull requests.
CursorSep 8Cursor has added Meta's Muse Spark 1.3 model to its AI code editor. The update includes new CursorBench 3.2 results, where the model's Max configuration achieved a 67.9% score at an average cost of $1.31 per task. Users can now select Muse Spark 1.3 for coding tasks directly within the editor.
CursorSep 3Cursor now supports running cloud agents on customer-managed infrastructure, keeping the agent loop in the Cursor cloud while moving execution to local networks or sandbox providers. This update enables agents to access internal services, specialized hardware, and custom build pipelines. Supported providers include AWS Lambda, Cloudflare, Modal, and Vercel, with auto-scaling pools available for enterprise-scale demand.
CursorSep 2Cursor added Google's Gemini 3.8 Flash model to its AI-first code editor. The update includes CursorBench 3.2 results, which plot the model's performance and cost relative to other available AI models. Gemini 3.8 Flash is now selectable for coding tasks within the editor.
CursorAug 22Cursor cloud agents now automatically pick up work from events like PRs or Slack threads and maintain a long-lived /goal until completion. The update adds custom modes for persistent skills, isolated virtual machines for subagents, and non-interruptive steering that queues follow-up messages for the next tool call. These changes allow agents to operate as autonomous, long-running systems.
CursorAug 22Cursor launched Origin, a code hosting platform now in beta for paid plans. The platform features repository hosting, pull requests, and real-time GitHub synchronization. Origin integrates with Vercel, Buildkite, and Depot to support preview deployments and CI workflows, with agent-native features designed for autonomous coding tasks arriving soon.
CursorAug 14Cursor is now officially part of SpaceX following today's acquisition close. The team joins xAI to work on Grok — including Grok Build, Grok Bot, Grok API, and Cursor — with the goal of making Grok the world's most useful AI.
CursorAug 13Cursor introduced builds, ready-to-use development environments prepared in the background. These builds allow cloud agents to start up to 3x faster by eliminating just-in-time setup. The system also improves reliability, as agents boot from the last successful build, ensuring work continues even if a recent commit or dependency update breaks the environment.
CursorAug 4Cursor open-sourced Mixture-of-Kittens, a deterministic training megakernel for mixture-of-experts models on NVIDIA GB300 NVL72 racks. The kernel fuses communication and computation, achieving up to 2.37x higher MXFP8 forward throughput than public baselines. In production, the implementation delivers a 1.41x increase in end-to-end training tokens per second across tens of thousands of GPUs.
CursorAug 3Cursor launched plugins that provide AI agents with direct read, write, and action capabilities across Google Workspace. These integrations connect agents to Gmail, Google Drive, Calendar, Docs, Sheets, and Chat. The plugins enable agents to search files, draft messages, manage calendar events, and edit documents directly from the editor using the Model Context Protocol.
CursorAug 3Cursor cloud agents now operate with 20–30% higher token efficiency and 80% greater efficiency during computer use tasks. These updates improve how agents manage MCP servers, skills, and graphical interface interactions. The changes allow for more ambitious task delegation while maintaining budget constraints.
CursorJul 30Cursor cloud agents now author 56% of merged pull requests in the company's internal monorepo, up from 10% in December. This shift follows the implementation of dedicated cloud computers for agents and a self-healing environment that monitors for setup failures. Agents now autonomously test changes end-to-end and perform root cause analysis to maintain their own development environments.
Frequently asked questions