The honest disclosure first
I use Claude Code all day and I built everything on this site with it. That is a bias, so here it is at the top rather than buried.
What this page does not do:
- No benchmark numbers. The public ones move every few weeks and none of them are measured on your work.
- No claim that one wins. The two most popular options here are not the same kind of thing, so "which is better" is the wrong question.
- No competitor prices I have not checked. The two below were read off the vendors' own pages on 2026-08-27.
Editor or agent, which is the real question
Almost every comparison in this space is really one distinction wearing different logos. Some of these products put the AI inside the place you work. Others put the work inside the AI.
- An editor with AI
- You are in the file. It suggests, completes and edits around your cursor. You accept or reject as you go. Cursor, and every ide with a copilot in it.
- An agent
- You are outside. You describe an outcome, it reads and changes files, runs commands, checks results, reports back. Claude Code, codex, Gemini CLI, opencode.
- How to tell which you want
- Do you want to watch every line change, or do you want to review the finished thing? That is the whole decision.
This matters because it predicts who is happy. People who like being in control of every line find agents unnerving. People who want their afternoon back find editors exhausting. Neither is wrong, and it is much easier to answer about yourself than to answer from a feature table.
Claude Code vs Cursor
| Claude Code | Cursor | |
|---|---|---|
| Shape | An agent in a terminal | An editor you work inside |
| Where you sit | Outside the work, reviewing results | Inside the file, accepting suggestions |
| Individual price | $20 a month on Pro | $20 a month on Pro |
| Free tier | none | Hobby, with limited agent requests |
| What else the plan includes | Claude Cowork, Claude design, Claude science | The editor, cloud agents, MCPs and hooks |
| Also runs in | Vs code, jetbrains, the web, a desktop app, slack, CI | Its own editor, plus a CLI |
| Good for non-code work | Yes. Folders, transcripts, spreadsheets, scripts. | Not really. It is built around a codebase. |
| The honest summary | Delegate a job, review the result | Write it yourself, faster |
Claude Code
- Shape
- An agent in a terminal
- Where you sit
- Outside the work, reviewing results
- Individual price
- $20 a month on Pro
- Free tier
- none
- What else the plan includes
- Claude Cowork, Claude design, Claude science
- Also runs in
- Vs code, jetbrains, the web, a desktop app, slack, CI
- Good for non-code work
- Yes. Folders, transcripts, spreadsheets, scripts.
- The honest summary
- Delegate a job, review the result
Cursor
- Shape
- An editor you work inside
- Where you sit
- Inside the file, accepting suggestions
- Individual price
- $20 a month on Pro
- Free tier
- Hobby, with limited agent requests
- What else the plan includes
- The editor, cloud agents, MCPs and hooks
- Also runs in
- Its own editor, plus a CLI
- Good for non-code work
- Not really. It is built around a codebase.
- The honest summary
- Write it yourself, faster
Prices read off Claude.com/pricing and Cursor.com/pricing on 2026-08-27. Both move.
The part that decided it for me
My work is not a codebase. It is folders of transcripts, lead spreadsheets, image pipelines and small scripts that call an API. An editor has almost nothing to offer that, because there is no file I want to sit inside. An agent has everything to offer it. If your work looks like mine, this is not a close call, and if your work is writing an application by hand, it might go the other way.
Claude Code vs Codex, Gemini CLI and opencode
These three are the same shape as Claude Code: an agent you talk to in a terminal that reads and writes files and runs commands. So the interface comparison people expect is mostly a non-comparison. What actually differs:
- The model behind it. The biggest real difference, and the one that moves most often. Whichever benchmark is winning this month will not be winning in three.
- The subscription you already have. If you pay for one of these ecosystems already, the marginal cost of its agent is usually zero, and that is a stronger argument than any feature.
- The ecosystem around it. Skills, sub agents, MCP servers, hooks, plugin marketplaces. This is where Claude Code is furthest along today, and it is the thing that compounds once you have written a few.
- Open source or not. Opencode is open, which matters if you need to self-host, audit it, or point it at your own model.
My honest recommendation: Do not choose on benchmarks. Choose on which model you already trust for your kind of work, and which subscription you already pay for. Then try the other one for a week in six months, because the ranking will have changed.
Claude vs Claude Code, which is a different question
A lot of the search traffic here is not a comparison at all, it is someone working out what the products are. Three names, one plan:
| Claude (chat) | Claude Cowork | Claude Code | |
|---|---|---|---|
| You get back | an answer | A finished document | A changed folder |
| Where it works | A chat window | Your files, apps and a built-in browser | A terminal, inside a folder |
| Best for | Thinking, drafting, questions | Decks, spreadsheets, reviews, web tasks | Code, scripts, pipelines, bulk file work |
| Included in Pro | yes | yes | yes |
Claude (chat)
- You get back
- an answer
- Where it works
- A chat window
- Best for
- Thinking, drafting, questions
- Included in Pro
- yes
Claude Cowork
- You get back
- A finished document
- Where it works
- Your files, apps and a built-in browser
- Best for
- Decks, spreadsheets, reviews, web tasks
- Included in Pro
- yes
Claude Code
- You get back
- A changed folder
- Where it works
- A terminal, inside a folder
- Best for
- Code, scripts, pipelines, bulk file work
- Included in Pro
- yes
All three come with the same $20 plan, so this is not a purchase decision, it is a habit decision. There is a fuller version of the cowork half on The cowork guide.
What Claude Code is genuinely worse at
A comparison page with no downsides is an advert. The ones I run into:
- Line-by-line control. If you want to see and approve every character, an editor is simply the right tool and Claude Code will feel like it is running away from you.
- The terminal is a real barrier. Not a technical one, a psychological one, and pretending otherwise does not help anyone. The desktop app and the VS Code extension exist partly for this.
- No free tier. Cursor lets you try the shape for nothing. Claude Code does not, so the first $20 is a bet.
- Long sessions drift. Usage and quality both degrade as a conversation grows. You have to build the habit of clearing between tasks, and nothing warns you.
- It will confidently do the wrong thing. So will all of them. The difference with an agent is that it does more of it before you look, which is why plan mode exists and why you read the diff.
How I would choose
In the order I would actually ask the questions:
- Is your work a codebase? If no, you want an agent, and most of this comparison collapses.
- Do you want to review lines or results? Lines means an editor. Results means an agent.
- What do you already pay for? The marginal cost of the agent in an ecosystem you are already in is usually nothing.
- Then stop reading and run both for a week. $20 each, and by friday you will know something no comparison page can tell you.
If you land on Claude Code, the setup and the first twenty minutes are on The Claude Code guide, the full cost picture is on The pricing page, and the CLAUDE.md habit is the one thing I would not skip.
Prices checked at claude.com/pricing and cursor.com/pricing on 2026-08-27. no prices are quoted for codex, gemini CLI or opencode because i did not verify them on the day, and a wrong price is worse than no price.