Based on 6 capability dimensions
8 versions
Cursor introduced multitasking with async subagents via /multitask command to parallelize requests, improved worktrees experience in the Agents Window for running isolated tasks across branches, and multi-root workspace support allowing a single agent session to target multiple folders for cross-repo changes.
Cursor 3.1 introduces interactive Canvases that allow the AI to respond with visualizations including dashboards and custom interfaces. The Agents Window received a tiled layout for running multiple agents in parallel, and voice input was upgraded with batch STT for higher-quality speech-to-text transcription.
Cursor added 30+ new plugins from partners like Atlassian, Datadog, GitLab, and Hugging Face, enabling cloud agents to read, write, and take actions across more of the development stack. Previous updates also introduced Automations for always-on agents triggered by schedules or events, JetBrains IDE support via Agent Client Protocol, MCP Apps with interactive UIs, and Bugbot Autofix for automatic PR fixes.
Cursor added 30+ new plugins from partners like Atlassian, Datadog, GitLab, and Hugging Face, enabling cloud agents to read, write, and take actions across more of the user's stack. Previous updates also introduced Automations for always-on agents triggered by schedules or events, JetBrains IDE support via Agent Client Protocol, MCP Apps with interactive UIs, and Bugbot Autofix for automatic PR fixes.
Composer 2 is now available in Cursor, offering frontier-level coding performance with strong results on challenging coding tasks. It introduces a two-tier pricing model: Standard at $0.50/M input and $2.50/M output tokens, and Fast (default) at $1.50/M input and $7.50/M output tokens. Additionally, self-hosted cloud agents are now supported, keeping code and tool execution entirely within the user's own network infrastructure.
13 versions
v3.80.0 adds enterprise-managed remote global skills with UI and system prompt integration, dynamic recommended models in onboarding, and a dedicated 'Quota Exceeded' error message. It fixes out-of-memory crashes during long conversations by increasing Node.js heap size to 8GB. It also removes foreground terminal mode, defaulting all task command execution to background mode.
Added Anthropic Opus 4.6 fast mode variants and fixed several issues including native tool placeholder interpolation in prompts, Windows unit test path normalization, and Bedrock thinking block handling. Markdown image loading now requires user consent, and hooks feature toggle was reintroduced.
Version 3.79.0 adds Claude Opus 4.7 model support and Azure Blob Storage as a storage provider. It fixes cache reflection for Cline and Vercel API handlers, addresses a stuck command_output issue when terminal commands end unexpectedly, and fixes an action injection security risk. The deprecated evals tool has been removed.
Added GPT-5.5 model support for OpenAI Codex subscription users. Removed hardcoded 'What's New' fallback items in webview so only remote-configured welcome banners are shown. Improved cline-core memory diagnostics including near-heap-limit heap snapshots, periodic memory usage logging, and heap snapshot discovery on abnormal exits for easier OOM debugging.
Version 3.75.0 brings latency improvements for remote workspaces. It also stabilizes flaky hooks tests and removes example hooks in favor of documentation-based guidance.