Choose Claude Code for Anthropic's Claude models in a terminal-first agent, and Codex if your team is on ChatGPT, wants OpenAI's GPT models or hands tasks to Codex Cloud. Both load SKILL.md skills and connect to MCP servers, so a team can run the two side by side, as ours does.[1][2][3][4][5][6]
Our team runs both. Of the 866 agent-written pull requests we merged from August 12 to October 4, 2026, Claude Code opened 822 and Codex 31, and both merged a median 37 minutes after opening.
Measure both with Zest:
Claude Code and Codex sessions side by side, per engineer
The model, skills and MCP tools behind each session
Claude CodeFable 5.1 (max) (with fallback)70.4 · Coding Agent Index
CodexGPT-6 Astra (max)67.0 · Coding Agent Index
Muse CodeMuse Spark 1.3 (xhigh)64.2 · Coding Agent Index
Grok BuildGrok 4.5 (high)64.1 · Coding Agent Index
Kimi Code CLIKimi K362.6 · Coding Agent Index
OpencodeGemini 3.8 Flash (high)61.2 · Coding Agent Index
Cursor CLIComposer 2.5 Fast38.3 · Coding Agent Index
Demo setups use current model families. The pairings above are the combinations benchmarked by Artificial Analysis; rankings measure performance, not popularity.
Illustrative daily trends · Other totals: last 30 days. Build Hours: this week, model-estimated activity including overlap; not engineer hours.
USED BY ENGINEERING TEAMS AT
AT A GLANCE
Claude Code vs Codex, side by side
What each tool documents, row by row.
Claude Code vs Codex, side by side
Claude Code
Codex
Team seats for 5 engineers
$100/mo — Claude Team Standard at $20 a seat, billed annually ($25 monthly).[7]
$100/mo — ChatGPT Business at $20 a seat, billed annually ($25 monthly).[3]
Where it runs
Terminal CLI, VS Code and JetBrains extensions, a desktop app, the web and mobile, plus Slack and GitHub Actions.[1]
Codex CLI, an IDE extension (VS Code, Cursor, Windsurf), the ChatGPT desktop app, Codex Cloud and @codex review on GitHub.[5]
Models
Claude models only (Fable, Opus, Sonnet, Haiku); Anthropic doesn't support routing Claude Code to other models.[2]
OpenAI GPT models; local clients can point at a custom Responses-API provider such as Ollama.[4][8]
Project instructions
CLAUDE.md and .claude/rules/; AGENTS.md only when there is no CLAUDE.md, or through an @import.[9]
AGENTS.md (global ~/.codex, then repo root down to the working directory) and AGENTS.override.md.[10]
Skills, subagents and hooks
Skills (SKILL.md, invoked as /skill-name), custom subagents in .claude/agents, and lifecycle hooks.[11]
Skills in .agents/skills (the Agent Skills standard), custom agents and subagents, and hooks.[6]
MCP servers
Local stdio or remote HTTP MCP servers, added with claude mcp add.[12]
STDIO and Streamable HTTP MCP servers in config.toml, from the app, CLI and IDE extension.[13]
Team analytics, built in
Team and Enterprise dashboard: daily users, sessions, lines accepted, accept rate and PRs with Claude Code; API on Enterprise.[14]
Workspace analytics of messages, credits and active users; a Codex dashboard and Analytics API on Enterprise.[15][16]
Checking your own usage
/usage (also /cost and /stats) shows plan usage, session cost and a 24h/7d breakdown by skill, subagent and MCP server.[17]
A usage dashboard with limits and reset times; /status and /usage daily, weekly or cumulative in the CLI.[3][18]
Each tool's column is read from its vendor's own docs and pricing page, dated in the sources below.
Claude Code or Codex: which one fits your team
Pick Claude Code if you want Claude. It runs Anthropic's models only (Opus, Sonnet, Fable and Haiku), from the terminal, VS Code, JetBrains, a desktop app, the web, Slack or GitHub Actions, and draws on the same usage pool as Claude chat.
Pick Codex if your team already pays for ChatGPT, since Plus, Pro and Business include it, or if you want OpenAI's GPT models, cloud tasks and @codex code review on GitHub. Its local clients can also point at another provider, such as Ollama, which Anthropic doesn't support for Claude Code.
For the heaviest users the limits decide it: Codex's Pro plan has no five-hour limit for now.
Claude Code vs Codex pricing and limits
Claude Code and Codex, from each vendor's pricing page in October 2026:
Claude Code comes with Claude Pro at $20 a month ($17 billed yearly), Max from $100 and Team seats at $25 ($20 billed yearly). Every plan's limits reset on a rolling five-hour window, and paid plans add weekly limits on top. More on Claude Code pricing.
Codex comes with ChatGPT Plus at $20 a month, Pro from $100 and Business at $25 per user ($20 billed yearly), and Free and Go ($8) run GPT-6 Luna in the desktop app. On Plus and Standard Business, OpenAI estimates 5–45 local messages per five hours on GPT-6 Astra, 15–160 on GPT-6.1 Sol and 350–3,000 on GPT-6 Luna. More on Codex pricing.
What our team saw running both
Of the 2,620 pull requests merged in our two repos from July 5 to October 3, 2026, 133 (5%) came from Codex, from 4 engineers. In the 30 days to October 3 the Codex share rose to 9.9%.
The agent that opens a pull request also writes a session-analytics block on it. Of the 866 merged with one from August 12 to October 4, 2026, Claude Code wrote 822 and Codex 31, all in September. Both merged a median 37 minutes after opening, and Codex's were a little smaller: 576 lines changed to 691.
Limits bit on both: on one engineer's laptop, Claude Code showed a usage-limit notice on 16 of 29 active days from September 4 to October 3, 2026, and in September Codex's weekly window reached 100% on 7 days.
Is Codex or Claude Code better? What our AI reviewer found
Our own AI reviewer, IonWarp, gives every pull request in our two repos a first review and flags should-fix and blocking issues. Comparing Codex's pull requests with Claude Code's from the same engineer, repo and week (August 12 to October 4, 2026), Codex's were flagged 1 point more often, with a 95% confidence interval from 16 fewer to 20 more: no measurable difference.
Pull request size mattered far more. After review, Codex declined 22 of the 58 findings it answered (38%), against 18% for Claude Code with the same engineers (August 31 to September 20); with 31 Codex pull requests, that is thin.
One small team, agent-reported names and a reviewer that changed week to week: how we collected it.
Merged pull requests in our two repos, August 12 to October 4, 2026: share flagged should-fix or blocking on IonWarp's first review.
Pull requests
Count
Flagged on first review
Codex vs Claude Code, same engineer, repo and week
24 vs 162
+1 point (−16 to +20)
Under 100 lines changed
298
11%
300 to 999 lines
334
45%
1,000 to 2,999 lines
185
57%
Running Claude Code and Codex on one repository
Keep shared instructions in AGENTS.md and import it from CLAUDE.md, so both agents read one file. Claude Code loads skills from .claude/skills and Codex from .agents/skills, in the same SKILL.md format, so a skill written for one can be copied to the other.
TRY IT ON YOUR TEAM
See what your team ships with AI.
Starter is free (3 seats, no credit card). Pro is $49 a month with 5 seats and Max is $149 a month with 10 seats — priced per workspace, not per engineer, with extra seats at $19 a month.
Winding / Teams
