Permanent editorial comparison

Cursor vs OpenAI Codex vs Claude Code

AI developers and engineering leads choosing a primary coding-agent workflow

RepoRadar conclusion

Choose by workflow, evidence, and operating model

Choose Cursor for an editor-native daily loop, Codex for parallel local/cloud delivery and review workflows, and Claude Code for a terminal-first agent with strong repository tool use. The best choice depends more on workflow surface, governance, and usage economics than on a single benchmark.

Decision matrix

Verification
Editorially reviewed
Installed and tested
Installed and tested
Pricing
Free: $0 month; Pro: $20 month; Pro Plus: $60 month; Ultra: $200 month; Teams: $40 user/month
Eligible ChatGPT plans: plan-dependent inclusion; Flexible Codex credits: token-based credits
Pro monthly: $20 month; Max 5x: $100 month; Max 20x: $200 month; Team annual: $25 user/month; Team monthly: $30 user/month
Deployment
desktop, vendor_cloud
local_cli, IDE_extension, desktop_app, vendor_cloud
local_cli, vendor_cloud, AWS Bedrock, Google Vertex AI
Privacy
Privacy behavior and code-data handling depend on account and Privacy Mode settings; teams should verify organizational controls before using proprietary repositories.
Code and task context are processed according to the selected ChatGPT, Business, Enterprise, API, or cloud-task terms; organizational controls should be reviewed before sensitive repository use.
Claude Code sends selected prompts and repository context to the configured model service; retention and training treatment vary by consumer, API, Team, Enterprise, Bedrock, or Vertex terms.
Extensibility
editor extensions, rules, MCP
skills, plugins, MCP, automations, hooks, SDK
MCP, hooks, subagents, custom commands
Best for
AI developers, technical founders, engineering teams
software teams, parallel repository work, review and automation workflows
terminal-first developers, large codebase work, engineering teams

Verification evidence

Evidence stages are separate from scores. Dated hands-on claims link to retained logs; records without a run say so explicitly.

OpenAI Codex

Installed and tested · last checked 2026-07-13T03:19:54.905224Z

Hands-on setup or capability check retained by RepoRadar verification harness

partial · cohort-20260712-codex-installed-help

Tester
RepoRadar automated local verification harness
Started
2026-07-13T03:19:54.809478Z
Completed
2026-07-13T03:19:54.905224Z
Environment
Windows 10 AMD64; Python 3.11.9; credential-stripped child environment; disposable home/cache
Install/setup time
0 minute(s)
Evidence scope
Bounded setup or capability check
Cleanup
Per-check temporary home and work directory removed. Shared cohort package cache removed.
Actions exercised
  • Created a disposable home, work directory, and isolated package cache with credential-like environment variables excluded.
  • Executed bounded check: Load the installed Codex CLI command surface.
  • Captured the complete sanitized stdout, stderr, exit status, and 0.10-second wall time.
Observed results
  • Command exited 0 after 0.10 seconds.
  • The command produced a successful local process exit; complete output is retained.
Observed strengths
  • The installed Codex CLI help exposed interactive prompting, non-interactive exec and review commands, and explicit login and logout controls.
Friction
  • No blocking friction was observed during this bounded local check.
Limitations
  • Help output proves local CLI startup only; it does not exercise an authenticated coding task.
  • This bounded cohort check is not a production benchmark or a claim of real user-workflow adoption.

Pricing assessment: No paid plan or metered provider usage was exercised; package or licensing, hosting, and provider costs remain workflow-dependent.

Privacy assessment: No repository content, user data, or provider prompt was transmitted; broader product data handling was not assessed by this bounded run.

Open retained test log →

Open full decision record →

Claude Code

Installed and tested · last checked 2026-07-13T03:20:06.168922Z

Hands-on setup or capability check retained by RepoRadar verification harness

partial · cohort-20260712-claude-installed-version

Tester
RepoRadar automated local verification harness
Started
2026-07-13T03:19:54.906223Z
Completed
2026-07-13T03:19:55.037759Z
Environment
Windows 10 AMD64; Python 3.11.9; credential-stripped child environment; disposable home/cache
Install/setup time
0 minute(s)
Evidence scope
Bounded setup or capability check
Cleanup
Per-check temporary home and work directory removed. Shared cohort package cache removed.
Actions exercised
  • Created a disposable home, work directory, and isolated package cache with credential-like environment variables excluded.
  • Executed bounded check: Read the installed Claude Code CLI version.
  • Captured the complete sanitized stdout, stderr, exit status, and 0.13-second wall time.
Observed results
  • Command exited 0 after 0.13 seconds.
  • The command produced a successful local process exit; complete output is retained.
Observed strengths
  • The installed Claude Code binary launched in 0.13 seconds and reported version 2.1.205 in the credential-stripped check.
Friction
  • No blocking friction was observed during this bounded local check.
Limitations
  • No authenticated prompt, repository edit, hook, MCP connection, or model call was attempted.
  • This bounded cohort check is not a production benchmark or a claim of real user-workflow adoption.

Pricing assessment: No paid plan or metered provider usage was exercised; package or licensing, hosting, and provider costs remain workflow-dependent.

Privacy assessment: No repository content, user data, or provider prompt was transmitted; broader product data handling was not assessed by this bounded run.

Open retained test log →

partial · cohort-20260712-claude-installed-help

Tester
RepoRadar automated local verification harness
Started
2026-07-13T03:19:55.037759Z
Completed
2026-07-13T03:19:55.316621Z
Environment
Windows 10 AMD64; Python 3.11.9; credential-stripped child environment; disposable home/cache
Install/setup time
0 minute(s)
Evidence scope
Bounded setup or capability check
Cleanup
Per-check temporary home and work directory removed. Shared cohort package cache removed.
Actions exercised
  • Created a disposable home, work directory, and isolated package cache with credential-like environment variables excluded.
  • Executed bounded check: Load the installed Claude Code CLI command surface.
  • Captured the complete sanitized stdout, stderr, exit status, and 0.28-second wall time.
Observed results
  • Command exited 0 after 0.28 seconds.
  • The command produced a successful local process exit; complete output is retained.
Observed strengths
  • Claude Code's retained help exposed interactive and --print non-interactive modes, additional-directory access controls, and agent selection.
Friction
  • No blocking friction was observed during this bounded local check.
Limitations
  • Help output proves local CLI startup only; it does not exercise an authenticated coding task.
  • This bounded cohort check is not a production benchmark or a claim of real user-workflow adoption.

Pricing assessment: No paid plan or metered provider usage was exercised; package or licensing, hosting, and provider costs remain workflow-dependent.

Privacy assessment: No repository content, user data, or provider prompt was transmitted; broader product data handling was not assessed by this bounded run.

Open retained test log →

partial · cohort-20260712-claude-isolated-npm

Tester
RepoRadar automated local verification harness
Started
2026-07-13T03:20:02.139885Z
Completed
2026-07-13T03:20:06.168922Z
Environment
Windows 10 AMD64; Python 3.11.9; credential-stripped child environment; disposable home/cache
Install/setup time
0 minute(s)
Evidence scope
Bounded setup or capability check
Cleanup
Per-check temporary home and work directory removed. Shared cohort package cache removed.
Actions exercised
  • Created a disposable home, work directory, and isolated package cache with credential-like environment variables excluded.
  • Executed bounded check: Resolve and launch the current Claude Code npm package in an isolated cache.
  • Captured the complete sanitized stdout, stderr, exit status, and 4.03-second wall time.
Observed results
  • Command exited 0 after 4.03 seconds.
  • The command produced a successful local process exit; complete output is retained.
Observed strengths
  • The pinned @anthropic-ai/[email protected] package resolved in an isolated npm cache, launched, and reported Claude Code 2.1.207.
Friction
  • No blocking friction was observed during this bounded local check.
Limitations
  • Version startup does not exercise authentication, repository access, or a model-backed workflow.
  • This bounded cohort check is not a production benchmark or a claim of real user-workflow adoption.

Pricing assessment: No paid plan or metered provider usage was exercised; package or licensing, hosting, and provider costs remain workflow-dependent.

Privacy assessment: No repository content, user data, or provider prompt was transmitted; broader product data handling was not assessed by this bounded run.

Open retained test log →

Open full decision record →

Dated first-party fact sources

Pricing, privacy, deployment, extensibility, and verification sources are merged from each product record. Each source keeps its own checked date; follow the official links before purchasing or standardizing.