Cursor alternatives in 2026, as OpenAI models leave Cursor

Cursor alternatives are worth pricing this month, even for teams that stay. As of 19 October 2026, OpenAI’s proposed 12 November shutoff for its models in Cursor is still unconfirmed. By Cursor’s count, OpenAI models serve about 5% of its traffic, and Cursor says its Anthropic, Google, Meta and Grok models continue.¹²³ The case for moving is strongest for teams that need OpenAI’s GPT-6 models, which Cursor never received, and for new buyers whose work runs mostly on Claude or Gemini. Cursor’s plans under The X Subscription, launched on 12 October, charge API prices for those models from the first request.⁴⁵

Claude Code and OpenAI’s Codex are the agent-first alternatives, and both can import a Cursor setup.⁶⁷ GitHub Copilot, Devin Desktop (Windsurf’s new name) and Zed put GPT-6 and Claude 5.5 models in one editor.⁸⁹¹⁰ JetBrains AI with Junie suits teams in JetBrains IDEs, Google’s Antigravity starts free, and open-source agents such as Cline, OpenCode and Pi run on your own keys.

This post covers what is confirmed about Cursor and OpenAI, the limits of an OpenAI key inside Cursor and who should switch. It then compares ten alternatives and ends with a decision rule and a migration checklist. Facts come from vendors’ own pages and from reporting dated on or before 19 October 2026, and we haven’t benchmarked the tools.

Cursor and OpenAI models in November: what is confirmed

SpaceX completed its purchase of Cursor on 14 August 2026.¹¹ On 28 August, OpenAI told SpaceX that it intended to wind down its contract supplying models to Cursor, with a proposed shutoff date of 12 November 2026, which it said was the maximum notice its contract allowed. It also said it would provide no future models to Cursor. OpenAI’s stated reason is that it cannot be confident SpaceX will use its technology within its terms of service.⁴

Nothing had been settled by 19 October. OpenAI’s help article on using its models in Cursor, last updated in early October, still calls 12 November a proposed date. It says Cursor may end access sooner and that OpenAI will share the official termination date once both companies confirm it.¹ Cursor’s chief executive, Michael Truell, said in August that OpenAI models serve about 5% of Cursor’s user traffic and that Cursor was talking to OpenAI.² On 29 August a Cursor support engineer wrote on Cursor’s forum that models from Anthropic, Google, Meta and Grok would continue through direct partnerships.³ Asked on 13 October whether newer OpenAI models might reach Cursor, he replied that Cursor had no information to share beyond what OpenAI had already said.¹² Anthropic co-founder Tom Brown said Anthropic would keep adding compute for Claude models in Cursor.²

Cursor’s model list on 19 October shows what is at stake.⁵

The X Subscription changed what Claude and Gemini cost in Cursor

For many Cursor users the bigger change came on 12 October, when SpaceXSI, the SpaceX company behind Grok, launched The X Subscription, a single plan for Grok, Grok Bot, Cursor and X Premium.¹³ New customers now start on it. Its paid tiers include a weekly pool of usage for Cursor’s own models, Grok and Composer, shared with the Grok apps. Claude, GPT and Gemini sit outside that pool. On Premium and above they are billed at API list price from prepaid credit from the first request, and the $8 Plus tier offers neither third-party models nor your own API keys.⁵¹²

Existing subscribers keep their legacy plans, and Pro, Pro+, Ultra and Teams include a monthly allowance for third-party models. Cursor maps each legacy plan to a comparable new tier. Prices are monthly, before tax, as of 19 October 2026.¹⁴

Switching is optional and permanent, and old plans can’t be bought again once you leave them.¹⁴¹² Cursor has given no timeline for moving everyone else, but its support engineer wrote that users who don’t switch “will, sometime in the future, be migrated with advance notice”.¹²

The New Stack worked through the cost for a Claude user. Pro included $20 of third-party usage at API prices, and Pro+ and Ultra included $70 and $400. A developer who used $20 of Claude a month paid $20 on Pro, and would pay $50 on Premium, the $30 plan plus $20 at API prices. SpaceXSI’s Matt Palmer told users that the new plans might cost more if most of their usage is on third-party models, which he said is one reason they are opt-in.¹⁵ In the other direction, SpaceXSI says every new plan includes at least twice as much coding usage with its own models as the comparable Cursor plan.¹³

Your own OpenAI key in Cursor, and its limits

OpenAI’s help article lists three ways to keep its models inside Cursor after the cut-off. You can add your own OpenAI API key for Cursor’s local Chat and Agent, install the Codex IDE extension, or route requests through an AI gateway such as Amazon Bedrock or Azure.¹ Each works, within limits that matter to a team.

• Local features only. Your key or gateway applies to Cursor’s local Chat and Agent. OpenAI’s article says it doesn’t apply to Cursor Tab, Auto, cloud or background agents, Automations, the Cursor CLI, or Cursor’s API and SDK.¹
• Plan. Bring-your-own-key works on Premium and above and on the legacy Pro, Pro+ and Ultra plans. It isn’t available on Plus.¹⁶¹²
• Which models. Cursor’s own page describes its OpenAI key support as “Standard, non-reasoning chat models” and says the model picker shows which are available. Check that the model you depend on appears with your key before the cut-off.¹⁶
• Prompt and tools. A Cursor support engineer explained in September that a custom model gets Cursor’s GPT-specific prompt and editing tools only if its name contains a built-in id such as gpt-5.6-sol. Cursor hasn’t said what happens to those built-in entries after the cut-off.³
• Data and routing. Requests made with your key still pass through Cursor’s servers for prompt building, and Cursor’s zero data retention policy doesn’t cover them.¹⁶
• Team cost. On Teams and Enterprise, Cursor still charges its Cursor Token Rate of $0.25 per million tokens on requests made with your own key.¹⁶

The Codex extension is the surest route to OpenAI’s newest models, because it runs OpenAI’s own agent on a ChatGPT plan or an API key. It is a separate panel, though, and it doesn’t change the model behind Cursor’s Chat, Agent, Tab or cloud agents.¹

Outside Cursor, OpenAI’s Sign in with ChatGPT lets subscribers spend their plan’s usage in participating tools and set a weekly limit for each app. As of 19 October the coding tools on OpenAI’s partner list include Amp, Devin, Kilo Code and Warp, plus the open-source OpenCode and Pi. Cursor isn’t on the list.¹⁷

Who should switch from Cursor, and who shouldn’t

Nothing forces a move on 12 November for a team that works mainly in Claude, Gemini, Grok or Composer. Stay, for now, if one of these fits.

• You’re on a legacy plan and mostly use Claude, Gemini or Grok. Those models continue, your included allowance for third-party models continues, and Cursor has said it will give advance notice before moving you.¹⁴¹²
• You mostly use Grok or Composer. The X Subscription is built around them, and SpaceXSI says each new tier includes at least twice the coding usage of the plan it replaces.¹³
• You use GPT models now and then. Your own key or the Codex extension covers local work.¹

Switch, or add a second tool, if one of these fits.

• Your team needs GPT-6 Astra or GPT-6.1 Sol. OpenAI won’t supply new models to Cursor, and the Codex extension reaches them only inside its own panel.⁴¹
• You want OpenAI models in cloud agents, Tab or the CLI, or under Cursor’s zero data retention. Your own key reaches none of those.¹¹⁶
• You’re buying new seats for mostly Claude or Gemini work. On The X Subscription that usage is billed at API prices on top of the plan, so price it against plans that include those models.⁵

Cursor alternatives compared on models, price, AGENTS.md and MCP

The table covers each tool as of 19 October 2026, with monthly US prices before tax. Cursor is included for reference.

Sources are each vendor’s pricing pages and documentation.^{5181962021228232492526272810293031323334}

Two large 2026 surveys show how widely the tools are used. They ask different questions, so compare rows within a column.

JetBrains surveyed more than 15,000 professional developers and reweighted the results; it also sells one of the tools it measured.³⁵ Stack Overflow’s shares are of the 12,255 respondents who answered the question in a survey that ran from June to August 2026, with results published on 6 October.³⁶

Claude Code and Codex, the agent-first alternatives

Claude Code is Anthropic’s coding agent, in the terminal, VS Code and JetBrains extensions, a desktop app, and cloud sessions that keep running after you close the laptop. Anthropic says its VS Code extension also works in Cursor and Devin Desktop.³⁷⁶ It runs Claude models only, with Opus 5.5 the default on every paid plan, and Claude Pro, Max, Team and Enterprise plans include it.¹⁹ Its /import cursor command brings Cursor’s instruction files, MCP servers, commands, subagents and skills across, and /init reads .cursor/rules/ and .cursorrules when it writes a CLAUDE.md.⁶ Code Review for pull requests became generally available on Team and Enterprise plans on 15 October.³⁸ For a Cursor user who mostly chose Claude, it is the most direct replacement for the agent, and a Claude plan includes Claude usage up to its limits where Cursor’s new plans charge API prices.

Codex is OpenAI’s agent, in the ChatGPT desktop app, a CLI, an IDE extension for VS Code, Cursor and Windsurf, JetBrains and Xcode integrations, and OpenAI’s cloud.²¹ It runs OpenAI’s models, led by GPT-6 Astra, which OpenAI calls its most capable, and GPT-6.1 Sol, which it recommends for complex coding. ChatGPT plans from Plus upwards include it, and Free and Go users get GPT-6 Luna in the desktop app as it rolls out.²¹²⁰ The desktop app and the CLI import from Cursor, turning instruction files into AGENTS.md and MCP settings into Codex’s own configuration.⁷ Because the same extension runs inside Cursor, Codex is the shortest route to GPT-6 for a team that wants to keep Cursor open.

Our comparison of Claude Code, Codex and Cursor covers independent benchmark results for the models behind both agents.

Cursor vs GitHub Copilot and Cursor vs Windsurf

GitHub Copilot works in VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse and Zed, in Copilot CLI and on GitHub.²² Its models come from OpenAI, Anthropic, Google and xAI, and Pro+, Max, Business and Enterprise get the newest, including GPT-6 Astra, GPT-6.1 Sol, Claude Opus 5.5 and Gemini 4 Argon.⁸ Since 1 June, usage has been metered in GitHub AI Credits, worth $0.01 each and spent at each model’s API rate. Code completions and next edit suggestions don’t use credits.³⁹ Business and Enterprise pool credits across the organisation, and extra usage is on by default until an admin turns it off.⁴⁰

Assign an issue to Copilot’s cloud agent and it works in an environment powered by GitHub Actions, opens a pull request and asks you for a review. Paid plans can also hand tasks to Claude and Codex as third-party agents, in public preview.⁴¹ Copilot reads AGENTS.md (the nearest file wins), a root CLAUDE.md or GEMINI.md, and .instructions.md files with an applyTo glob, which map closely onto Cursor’s file-scoped rules.⁴² For a team that already buys GitHub, it keeps a multi-vendor model picker, now with GPT-6, under one seat price.

Windsurf is now Devin Desktop. Cognition, which signed its deal for Windsurf in July 2025, renamed it on 2 June 2026 and describes it as the same IDE with the same features.⁴³⁴⁴ Every plan has unlimited Tab completion, extra usage is billed at API prices, and Pro, Max and Teams include Devin cloud sessions, each on its own virtual machine.²⁴⁹ Its models include Cognition’s SWE-2 alongside GPT-6 Astra, GPT-6.1 Sol, Claude Opus 5.5 and Gemini 4 Argon.⁹ It reads AGENTS.md, keeps .windsurf/rules/ working beside its new .devin/rules/, imports .cursor/rules and can import a Cursor configuration.⁴³⁹ Anthropic lists it among the editors that run its VS Code extension.³⁷ Of the tools here it is the closest in shape to Cursor, with its own completion model, several labs’ models and cloud agents in one product.

JetBrains AI, Zed, Antigravity and the open-source agents

JetBrains AI sells credits worth $1 each, and code completion doesn’t use them.²⁵⁴⁵ Paid tiers run Junie, Claude Agent, Codex and Gemini, and JetBrains lists more than 25 agents, Cursor’s among them, that connect through the Agent Client Protocol (ACP) with your existing accounts.²⁵ Models include Claude Opus 5.5 and GPT-6.1 Sol, and Junie can call GPT-6 Astra.²⁶ Junie reads AGENTS.md, and IntelliJ IDEA’s first-run import brings Cursor’s keymap, theme, extensions and recent projects across.⁴⁶⁴⁷ Cloud agents run through JetBrains Air, for organisations only.²⁷

Zed is a native editor for macOS, Linux and Windows. It is free with your own keys or external agents, and Pro adds unlimited edit predictions and hosted models, billed at API price plus 10% after the first $5 a month.²⁸¹⁰ Through ACP it runs Claude, Codex, Copilot, Cursor, OpenCode and Pi as external agents without charging for them, and it offers a Cursor base keymap.⁴⁸⁴⁹ For project instructions it uses the first file it finds from a list that starts with .rules and .cursorrules, so a leftover .cursorrules wins over AGENTS.md.²⁹

Google Antigravity is a desktop app for managing background agents, plus a CLI, an IDE and extensions for VS Code, Visual Studio, JetBrains IDEs, Zed and Xcode.³¹⁵⁰ It is free with basic weekly limits, and Google AI Pro or Ultra raises them.³⁰ Paid plans add Gemini 4 Argon and Claude Opus 5.5 and Sonnet 5.5 to Gemini 3.8 Flash, with no GPT-6 model, and Antigravity reads AGENTS.md or GEMINI.md in any folder.³¹

Open-source agents run on whichever provider you choose. Cline, under Apache 2.0, works in VS Code, JetBrains IDEs, the terminal and a desktop app. Individuals pay only for inference, and it reads AGENTS.md and a legacy .cursorrules file.³² OpenCode, under MIT, works with more than 75 providers and with ChatGPT Plus, GitHub Copilot and GitLab Duo subscriptions.³³ Pi, also MIT-licensed, is a minimal terminal harness that leaves features such as subagents to extensions.³⁴ Roo Code, often listed beside Cline, shut down on 15 May 2026, and its README points users to ZooCode, a community fork, and to Cline.⁵¹ Kilo Code, another MIT-licensed agent, offered Cursor Enterprise customers its Enterprise plan free for the rest of their Cursor contract, up to a year, on 16 October, with inference charged separately.⁵²

The best AI code editor in 2026 for each kind of team

Decide in this order.

1. Start with the models you must have. If they include GPT-6 Astra or GPT-6.1 Sol, that work moves off Cursor’s own agent. Use Codex, which also runs inside Cursor, or keep it in an editor with Copilot Pro+ or Business, Devin Desktop, Zed or JetBrains AI. If it’s mostly Claude, Claude Code on a Claude plan is the direct route. If it’s mostly Grok or Composer, Cursor remains the natural home.
2. Then match where your developers work. Terminal-first teams should trial Claude Code and Codex. Teams that like Cursor’s VS Code-based editor can go back to VS Code with Copilot or move to Devin Desktop. JetBrains users have JetBrains AI with Junie, plus Claude Agent, Codex and Cursor through ACP. Zed suits developers who want a fast native editor and run several agents side by side.
3. Then choose how you pay. Claude, ChatGPT, Copilot, Devin and JetBrains plans include usage up to a limit. Zed Personal, Cline, OpenCode and Pi cost nothing themselves and pass inference costs through to your provider. On a legacy Cursor plan with a third-party allowance, nothing changes until Cursor gives notice, so use that time to price the alternatives against Premium or Super plus API usage.

Run the same real tasks through your top two candidates for two weeks before you buy seats, with AGENTS.md in place so both read the same rules.

Migration checklist: rules, MCP, keybindings, seats and spend caps

1. Rules files. Move always-on rules from .cursor/rules/*.mdc into a root AGENTS.md, and folder rules into AGENTS.md files in those folders.⁵³ Rules scoped to file patterns need each tool’s own format, such as Copilot’s .instructions.md files with applyTo.⁴² Cursor’s User Rules and Team Rules live in its settings and dashboard, outside the repository, so copy them out by hand.⁵⁴ Claude Code loads AGENTS.md only when there’s no CLAUDE.md, so add an @AGENTS.md line to any CLAUDE.md you keep.⁶ If your team uses Zed, delete any leftover .cursorrules.²⁹
2. MCP config. Collect .cursor/mcp.json from each repository and ~/.cursor/mcp.json from each machine.¹⁸ A root .mcp.json with the same mcpServers key is read by Claude Code and by Copilot in VS Code, and VS Code can also discover Cursor’s servers.⁶²³ Claude Code needs a "type" such as "http" on remote entries. Codex uses TOML tables in config.toml, Zed uses context_servers, and Antigravity names the remote address serverUrl.^6215531 Sign in again to OAuth servers in the new client, and pin package versions while you’re in the file. Our post on agent config files that run code explains why.
3. Keybindings and settings. Cursor is built on VS Code’s codebase and uses VS Code’s default shortcuts, plus a few for its AI features such as Cmd+K for inline edit and Cmd+I or Cmd+L for the side panel. Note which of those your team relies on.⁵⁶⁵⁷ VS Code profiles carry settings, keyboard shortcuts, MCP servers and extensions in one .code-profile file.⁵⁸ IntelliJ IDEA imports Cursor’s keymap, Zed has a Cursor base keymap, and Devin Desktop can import a Cursor configuration.⁴⁷⁴⁹⁹
4. Team seats. Count who used Cursor’s agent in the last 30 days, and buy heavier seats only for heavy users. A cancelled legacy Cursor plan runs to the end of the period you’ve paid for, while switching to The X Subscription is permanent and refunds the unused time.¹⁴¹²
5. Spend caps. Set them before the first invoice. Copilot Business and Enterprise allow extra usage by default until an admin turns the policy off, and a $0 user budget blocks a user at once.⁴⁰ On The X Subscription, a monthly additional-usage limit caps top-up spending, and Zed Pro has its own spending cap.⁵²⁸ Our post on what coding agents cost per engineer shows which caps are hard stops and includes a policy template.

Frequently asked questions