Notes on coding agents and the pages they leave behind, on moving between notes apps, and on the plain mechanics of keeping what you think. Written and published from a Sumibako vault.
Claude Code vs Cursor in October 2026: after the SpaceX deal, the price change and auto mode
Claude Code vs Cursor in October 2026: what each is, what they cost now, how each rations Claude, and the $28 setup that runs one inside the other.
Prashant Shaw
Founder, Sumibako
9 min read
Every comparison of Claude Code and Cursor on the first page of a search today was written between February and September, and they disagree with each other about prices, model names and context windows. Since the newest of them, Cursor has become part of SpaceX, moved its plans into The X Subscription, started billing Claude at API list price, and retired Cmd+K. Claude Code has made auto mode the default, grown a desktop app, a browser version and a phone app, and most of its models now have a million-token window. So here is the comparison as of 29 October 2026, from both companies' own pages, by someone who uses both and sells neither.
🥊 The short answer
Claude Code
Cursor
What it is
An agent that runs in your terminal, VS Code, JetBrains, a desktop app, a browser and your phone
A code editor with an agent inside it, plus a CLI and cloud agents
Models
Anthropic's
Cursor's own Composer models and Grok on every plan; Claude, GPT and the rest at API list price from Premium up
What it costs
Included in Claude Pro at $20 a month, or Max at $100 or $200
Plus $8, Premium $30, Super $100, Ultra $200. The $20 Pro plan is closed to new buyers
From your plan's pool: a five-hour session limit and a weekly one
From a prepaid balance, at API list price, from the first token
Working while you are away
Auto mode, cloud sessions, scheduled tasks, Dispatch from your phone
Cloud Agents and automations from Premium up, and Grok Bot
Sharing what it made
Artifacts, on paid plans
Shared canvases
Runs inside the other
Yes: the Claude Code extension has an "Install for Cursor" link
No
The last row is the one most comparisons miss, and it changes the question. You do not have to choose an editor and an agent together.
🧰 One is a tool and the other is a room
Anthropic's one-line description is that "Claude Code is an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools." It has no editor of its own. It runs in your terminal, inside VS Code and JetBrains, in a desktop app that "includes Claude Code, so you don't need to install the CLI separately", in a browser at claude.ai/code, and from the Claude app on a phone. Anthropic's platforms page lists all six and says what most people find out by accident: "You can mix surfaces on the same project. Configuration, project memory, and MCP servers are shared across the local surfaces."
Cursor is a fork of VS Code with an agent built into the sidebar. Its own description: "Agent is Cursor's assistant that can complete complex coding tasks independently, run terminal commands, and edit code. Access in sidepane with Cmd+I." It has Tab completion, which Claude Code does not have at all, a Plan mode and an Ask mode, checkpoints that let you roll files back, and a CLI called agent for people who want the terminal version. For bigger jobs there are Projects, where "a coordinator agent plans the work and delegates it to other agents."
The practical difference is that Cursor is a place you work in and Claude Code is a thing you run, and because Cursor is VS Code underneath, you can run the second inside the first. Anthropic's VS Code page offers an "Install for Cursor" link next to the VS Code one. Half the people I know who "use Cursor" use it this way: Cursor's Tab for the small edits, Claude Code's panel for the work that takes an hour.
None. "Plus doesn't include third-party models like Claude, and it doesn't support using your own API keys"
Premium
$30
At API list price from a prepaid balance
Super
$100
Same
Ultra
$200
Same
Legacy Pro
$20, closed to new subscribers
Included usage, if you never switch
The sentence to remember is on Cursor's own page: "Premium and above include generous usage of Grok models but do not include an allowance for third-party models." And the one after it: "Third-party models remain available in Cursor at API list price." Claude Sonnet 5.5 in Cursor is $2 per million input tokens and $10 per million out, and Opus 5.5 is $4 and $20, which are Anthropic's numbers with no discount for being your editor. If you are on a legacy plan, nothing changes until you press the button, and "Switching to The X Subscription is optional and permanent."
Anthropic's side is one line on its pricing page: "Claude Code is included in all paid plans. It shares the same usage limits as the rest of your plan, so your work in the terminal and your chats draw from one pool." Pro is $20 a month, Max is $100 or $200 for five or twenty times the usage, and Max now also carries "$100 to $200/month in API credits to build on the Claude Platform". The catch is the pool: "Every plan has usage limits that reset on a rolling five-hour session window, and paid plans add weekly limits on top." I wrote up how those limits behave on a heavy day, because they are the thing you will hit.
So the two companies ration Claude in opposite ways. At Anthropic it is inside the subscription and you run out of time. At Cursor it is outside the subscription and you run out of money. The combination that gets the most Claude for the least is also the cheapest of the lot: Cursor Plus at $8 for the editor and Tab, plus Claude Pro at $20 with Claude Code installed inside it, $28 a month and two meters to watch.
📝NOTE
Cursor's shared pool has one property Claude's does not: "Your allowance refreshes weekly, so one heavy day never wipes out your whole month." Claude's five-hour window does the same job at a finer grain, and then adds a weekly cap on top. Neither is unlimited, whatever the plan name says.
🤖 Who is allowed to work while you are not looking
This is where the two have drifted furthest apart this year.
Claude Code's answer is a permission mode. Since version 2.1.283, "auto mode is the built-in starting permission mode for interactive terminal and VS Code sessions." In auto mode "a second model, the classifier, reviews actions instead of you", and only the things it refuses come back as a prompt. Add cloud sessions, which run on Anthropic's machines and keep going after you close the laptop, scheduled tasks in the desktop app, and Dispatch, which lets you message a job from your phone to a desktop at home, and the design is clear: one agent, many places, and a model standing between it and your shell.
Cursor's answer is more agents. Cloud Agents come with Premium and above, and the CLI can hand a conversation to one mid-sentence: "Push your conversation to a Cloud Agent to continue running while you're away." Automations schedule them. Projects put a coordinator over a team of them. And then there is Grok Bot, SpaceXAI's general-purpose agent that runs on Cursor's infrastructure and is "included with every paid individual Cursor plan and with the Cursor Teams plan", which is not a coding tool at all: each Bot gets its own cloud computer and logs into your other apps. If you want an agent to do your expenses, Cursor is now, oddly, the editor that comes with one.
Whether you trust either to run unattended is a separate question, and the only honest answer is that both companies say to use a sandbox.
🧠 Memory and context
Claude Code's models, the Fable models, Sonnet 5 and later, Haiku 5.5 and the recent Opus models, "support a 1 million token context window", and when a session gets near it the conversation is compacted into a summary. I wrote about what compaction keeps and loses this week; the short version is that files survive it and the conversation does not. Across sessions it remembers through CLAUDE.md, which you write, and auto memory, which it writes. Cursor's equivalents are rules, which load into every agent run, and checkpoints for the code, which "are stored locally and separate from Git." Both tools reload their instruction files at the start of every session, and both lose what you said in chat the moment the chat is gone.
If you have been in Cursor a while, one thing to know: on 26 October Cursor removed Cmd+K in the terminal. The staff answer on the forum was plain, "Cmd+K, including generating commands in the terminal, was retired today", and the replacement is the Agent on Cmd+I. People who had the shortcut in their fingers were not pleased, and I understood every word of it.
📤 What you get to keep
The output of an hour with either tool is code in your repository, and that part is the same. The difference is everything that is not code: the plan, the investigation, the explanation you want to send to a colleague.
Claude Code publishes artifacts, which "are available on Pro, Max, Team, and Enterprise plans": a page on claude.ai that updates while the session runs and can be shared or made public. Cursor has shared canvases, and Codex has Sites, and I compared the three in September. The plans Claude Code writes in plan mode are kept on disk and deleted after 30 days by default. Cursor's checkpoints are local to the machine. In both cases the durable things are the ones you deliberately wrote down somewhere else, which is a point I will keep making until it stops being true.
⚖️ Pick by how you work
If you...
Then
Live in a terminal, or want the agent in CI, on a server, or in a browser
Claude Code. The CLI is the most complete surface and Cursor's CLI is younger
Want Tab completion and an editor built around the agent
Cursor, on Plus if Grok and Composer are enough, Premium if you want cloud agents
Mostly use Claude models all day
Claude Pro or Max, and run Claude Code wherever you like, Cursor included
Want both and the least money out
Cursor Plus and Claude Pro, $28, with Claude Code installed in Cursor
Are on legacy Cursor Pro and use Claude in it daily
Stay there and do not press Unify
Want an agent for things that are not code
Cursor, because Grok Bot comes with it, or Claude's own apps outside Claude Code
Want to go away and come back to finished work
Either. Cloud sessions on one side, Cloud Agents on the other, and a sandbox on both
The honest summary is that they are less rivals than they were in February. Cursor is becoming a place where several agents, most of them not Anthropic's, do work for you, and Claude Code is becoming one agent that is available wherever you happen to be. The one thing I would not do is pay for Claude twice: once inside a Claude plan and again, by the token, inside Cursor.
🗂️ Wherever you end up
Whichever you pick, the plan the agent wrote before it touched your code, and the note it should leave at the end of the session, are worth keeping somewhere both of them can reach and you can open from a phone. I make Sumibako, a notes app that Claude Code, Codex and Cursor each file a page into with one command, with a link back you can hand to someone. It does not care which editor you use, and it has never asked anyone to pick a tier.
Cursor's pricing since SpaceX: the X Subscription tiers, what happened to the $20 Pro plan, what Claude costs in Cursor now, and when Claude Code is cheaper.
