Why this comparison matters
These three products overlap more now than their origins would suggest. Claude Code started as Anthropic's terminal agent. OpenAI Codex is OpenAI's coding agent, and it now comes with every ChatGPT plan. Cursor is a VS Code fork that has added an agent, cloud agents and a CLI on top of its editor. Each of them can take a task, edit several files and run commands. Where they differ is the place you drive the work from, the models you get, and how the bill behaves once you use it heavily.
This comparison draws on each vendor's pricing pages and docs and on the scored.tools ratings, which are built from published benchmarks and independent reviews. scored.tools does not test tools hands-on, so every performance claim below names its source. On our scores, Cursor leads with 7.8. Claude Code is close behind at 7.6, and Codex sits at 6.
Feature comparison
| Claude Code | OpenAI Codex | Cursor | |
|---|---|---|---|
| Vendor | Anthropic | OpenAI | Anysphere |
| Main way to use it | Terminal agent | Agent spread across ChatGPT surfaces | AI code editor built on VS Code |
| Other surfaces | VS Code, JetBrains, Slack, web | Desktop app, web, CLI, IDE extension, cloud environment, limited iOS app (per OpenAI) | CLI, cloud agents, Slack, Teams, Jira, Linear, Notion and GitHub integrations |
| Operating systems | macOS, Linux, Windows | Varies by surface and plan, per OpenAI | Desktop editor (VS Code base) |
| Models | Anthropic's Claude models | OpenAI GPT-6 family (Astra, Sol, Luna) | Several frontier models, including Claude, GPT and Grok |
| Free path | None. Claude's Free plan excludes Claude Code | Yes. ChatGPT Free includes Codex | Yes. Hobby plan, no card, limited agent requests |
| Cheapest paid plan | Pro, $17/month annual or $20 monthly | Go, $8/month | Individual, $20/month |
| Team plan | Team, from $20/seat/month annual | Business, $20/user/month | Teams, $40/user/month |
| Developer extensibility | GitHub, IDE and Slack integrations; Anthropic API | OpenAI API with per-token prices | TypeScript and Python SDK, MCP, skills, hooks |
| Published quality evidence | 80.8% on SWE-bench and 67% wins in blind tests, per NxCode's 2026 ranking | 64.6% on SWE-bench Pro and 88.8% on Terminal-Bench for the underlying model, per CodeAnt | 4.5/5 after six months from No Code MBA; #3 on a Coding Agent Index |
| scored.tools rating | 7.6 | 6 | 7.8 |
Be careful with the benchmark row. SWE-bench and SWE-bench Pro are different tests, and the Pro set is harder. Claude Code's 80.8% and Codex's 64.6% therefore can't be compared directly. Both are strong numbers. Neither proves one agent beats the other on your codebase.
Score breakdown
| Criterion | Claude Code | OpenAI Codex | Cursor |
|---|---|---|---|
| Output quality | 9 | 7 | 8 |
| Ease of use | 7 | Not scored (no evidence) | 8 |
| Pricing value | 6 | 5 | 7 |
| API and integrations | 8 | 7 | 8 |
| Problem fit | 8 | 5 | 8 |
| Overall | 7.6 | 6 | 7.8 |
Claude Code has the highest output quality score of the three, based on NxCode's ranking. Cursor wins on ease of use and pricing value because it has a free plan with no card required and docs that cover agents, rules, MCP and the CLI. Codex scores lower on problem fit because the evidence behind its multi-step workflow claims comes from OpenAI's own description, and no independent source confirmed it at scoring time.
Pricing comparison
Claude Code
Anthropic's pricing page lists Claude Code on Pro ($17/month billed annually, $20 monthly) and Max (from $100/month, with 5x or 20x Pro's usage). Team seats cost $20/month annual or $25 monthly for Standard, and $100 or $125 for Premium. Enterprise starts at $20/seat plus usage. The Free plan does not include Claude Code. Per Anthropic, Claude Code draws on the same usage limits as the rest of your Claude plan, so long chat sessions and long agent runs come out of one allowance. Fast mode, a research preview, bills separately at $8 per million input tokens and $40 per million output tokens.
OpenAI Codex
OpenAI's Codex pricing page says every ChatGPT plan includes Codex. The tiers are Free ($0), Go ($8/month), Plus ($20/month), Pro (from $100/month, with $100, $200 and $500 options), Business ($20/user/month), and custom pricing for Enterprise and Edu. The catch is how usage is counted. OpenAI quotes local message ranges per five-hour window for Plus and Standard Business: 5 to 45 messages on GPT-6 Astra, 15 to 160 on GPT-6.1 Sol and 350 to 3,000 on GPT-6 Luna. Ranges that wide make spend hard to predict, and that is why Codex scores 5 for pricing value. OpenAI says Pro plans currently have no five-hour limit. If you go through the API, GPT-6.1 Sol costs $2 in and $10 out per million tokens, and GPT-6 Luna costs $0.10 in and $0.50 out.
Cursor
Cursor's pricing page lists Hobby (free, limited agent requests, access to Composer) and Individual at $20/month, which covers Pro, Pro+ and Ultra tiers with higher agent limits. Individual adds frontier models, MCPs, skills and hooks, and cloud agents. Teams costs $40/user/month and adds SSO, team-wide privacy mode, shared rules and usage analytics. Enterprise is custom. Bugbot code review is usage-based on Individual. No Code MBA's review says the credit system can cost more than the sticker price, so treat $20 as the minimum you'll pay.
What a solo developer actually pays
At the entry level, the three cost about the same: $20 for Claude Pro monthly, ChatGPT Plus or Cursor Individual. Heavy users of Claude Code or Codex move to $100/month tiers. Codex is cheapest to try, at $0 or $8. Claude Code is the only one with no free route in. For teams, ChatGPT Business and Claude Team Standard both start at $20 a seat, while Cursor Teams costs twice that.
Use case scenarios
Pick Claude Code for large, agent-driven changes
Claude Code is built for handing off a whole task, such as a migration, a bug fix or a test suite, and then reviewing the result. Anthropic says it handles multi-day migrations, and you can steer it from the terminal, your IDE, Slack or the web. It has the best published quality evidence here: 80.8% on SWE-bench per NxCode. The Pragmatic Engineer survey, as cited by Gradually.ai, found that 46% of respondents named it their most loved tool. If you mainly work in the terminal and want the agent to do the bulk of the work, this is the pick. Budget for Max if you plan to run it all day.
Pick Cursor for editing inside your IDE
Cursor suits developers who want the AI built into the editor they already know. It runs on VS Code, so extensions and keybindings carry over. Tab completion, codebase chat and multi-file edits sit next to an agent mode. It is also the only one of the three that lets you switch between Claude, GPT and Grok models, which helps if you don't want to be tied to one lab. Teams get SSO, privacy mode and a TypeScript and Python SDK. It also has the best entry path, a free plan with no card required.
Pick Codex if you already pay for ChatGPT
If your team already pays for ChatGPT Plus or Business, Codex costs nothing extra to try. It reaches more surfaces than either rival, including a desktop app, the web, the CLI, an IDE extension, cloud tasks and a limited iOS app. The underlying model's 88.8% on Terminal-Bench, per CodeAnt, suggests it handles shell-heavy work well. On the other hand, our evidence on setup and day-to-day ease of use is thin, and its workflow claims have not been checked independently. Try it on a real task before you standardise on it.
Pick Codex for the cheapest start
Codex is the only one of the three that includes an agent on a $0 plan from a frontier lab, and Go at $8 is the cheapest paid tier on this page. Students and hobbyists should start here. Cursor Hobby comes second.
Verdict
Cursor has the highest overall score at 7.8. It gets there through balance: a free plan, a familiar editor, a choice of models and good team controls, with no weak criterion. Most developers who live in an IDE should start with Cursor.
Claude Code, at 7.6, wins on output quality. Its 9 is the highest single criterion score in this comparison. For delegating large refactors, migrations and bug hunts to an agent, it is the strongest choice, as long as you accept that it has no free tier and that heavy use means the $100 Max plan.
Codex, at 6, has a capable model and the widest range of surfaces and price points. The trouble is that its usage limits vary so much that costs are hard to predict, and its workflow claims still rest on OpenAI's own word. It makes sense for teams already on ChatGPT and for anyone who wants an agent for free. For everyone else, it is a second choice for now.
