AI Agent vs Chatbot: What's the Real Difference?
AI agent vs chatbot compared on control flow, tool use and autonomy, with sourced definitions from Salesforce, Google Cloud and Wikipedia.
6 min read
Claude Code vs Codex compared on pricing, GitHub automation, open source status and the latest releases from Anthropic and OpenAI.
Claude Code is Anthropic's agentic coding tool, bundled into Claude subscription plans starting at $17 a month. Codex is OpenAI's coding agent, included in ChatGPT plans and built on an open source CLI. Claude Code leads on configuration depth and parallel sub-agents. Codex leads on GitHub pull request automation and is the one you can actually read the source code of.
Both tools read a codebase, edit files across it, run shell commands, and work with git to stage changes, write commit messages, and open pull requests. The claude code vs codex choice usually comes down to three things: which company makes the model you already trust, which subscription you already pay for, and whether GitHub automation or deep terminal configuration matters more to how your team works.
The table below lists the differences that actually change a buying decision, pulled from each vendor's own documentation and pricing pages.
| Aspect | Claude Code | Codex |
|---|---|---|
| Maker | Anthropic | OpenAI |
| Interfaces | Terminal CLI, VS Code, JetBrains, desktop app, web (claude.ai/code), mobile | ChatGPT app, Codex CLI, IDE extension, Codex cloud |
| Source | Closed source, ships as the npm package @anthropic-ai/claude-code | Open source CLI and SDK under the Apache-2.0 license |
| Instructions file | CLAUDE.md, and AGENTS.md when no CLAUDE.md is present (added September 19, 2026) | AGENTS.md |
| Parallel work | Sub-agents and background agents coordinated by a lead agent | Multi-agent workflows using computer and browser tools, plus a reusable cloud environment per repo |
| GitHub automation | GitHub Actions integration, code review | GitHub App that auto-reviews pull requests and replies to @Codex mentions on issues |
| Entry-level price | Free tier, then Pro at $17 a month billed annually or $20 monthly | Included in ChatGPT plans; shown as €23 a month for Plus on OpenAI's own pricing page |
| Top individual tier | Ultra at $200 a month, no 5-hour usage limit | ChatGPT Pro, referenced at $200 and, in OpenAI's own rate card, a newer $500 a month tier for Ultrafast access |
Claude Code is the agent Anthropic built to live in a developer's existing tools rather than a separate app. It runs as a terminal CLI, a VS Code or JetBrains extension, a standalone desktop app, a browser session at claude.ai/code, and on the Claude mobile apps, all tied to one Claude account. It needs a Claude subscription or an Anthropic Console account; the terminal CLI, VS Code and JetBrains versions also support third-party model providers.
Claude Code reads a CLAUDE.md file at the start of every session for a team's coding standards and architecture decisions. Since September 19, 2026, it also reads a repository's AGENTS.md file in place of CLAUDE.md when no CLAUDE.md exists, closing a gap that previously forced teams to keep a separate instructions file just for Claude. It builds "auto memory" as it works, too, saving what it learns across sessions without anyone writing it down.
The tool supports skills for packaging repeatable workflows, hooks for running shell commands around its own actions, and sub-agents that split one task across several Claude Code instances coordinated by a lead agent. Background agents let a developer run several full sessions at once and watch them from a single screen, and the Agent SDK lets teams build fully custom agents on top of Claude Code's own tools and permissions.
Anthropic ships the CLI fast. Its public changelog lists version 2.1.328 on November 19, 2026, with three more releases in the week before it. The GitHub repository anthropics/claude-code holds Claude Code's plugins, examples and issue tracker, but the CLI itself is a closed source npm package rather than a public source tree.
Codex is OpenAI's coding agent, included in a ChatGPT plan and built to behave the same way across the ChatGPT app, a browser-based Codex cloud, an IDE extension, and the Codex CLI, all tied to one ChatGPT account. OpenAI's own product page lists Plus at €23 a month, Pro at €103 a month, and Business at €21 per user per month for two or more seats billed annually.
Codex is built around multi-agent workflows. It can run many agents in parallel, using computer and browser tools to verify its own work, and it keeps cloud tasks running after a laptop closes, with a reusable cloud environment for every connected repository. Its GitHub App automatically reviews pull requests, comments inline on specific lines, and responds whenever a user tags @Codex on an issue or a pull request thread.
Codex's source, unlike Claude Code's, is public. The openai/codex repository on GitHub carries an Apache-2.0 license, 135.8k stars, and 21.5k forks, and its latest tagged release was 0.170.0 at the time of research. OpenAI's own ChatGPT rate card, last updated around November 18, 2026, says a typical Codex task on its current GPT-6.2 Sol model costs between 2 and 15 credits, and that Codex usage for a business or enterprise developer typically runs $100 to $200 a month under token-based billing, separate from the flat ChatGPT subscription price. The same rate card confirms GPT-5.3-Codex is the model behind Codex's automated pull request review.
Which tool wins depends more on what a team already pays for and automates than on raw model quality. A developer already on a Claude subscription who wants deep configuration, custom hooks, slash commands, and sub-agents working in parallel gets more from Claude Code without adding a second tool. A team that leans on GitHub for code review and wants an agent that reacts to @-mentions on issues and pull requests gets more from Codex's GitHub App, a workflow Builder.io's Steve Sewell rated above Claude Code's GitHub integration in his own test of both tools. A team that wants to read, fork, or audit the agent's own source code has one real option, Codex, since Claude Code is not open source. Teams already running ChatGPT Business or Enterprise credit-based billing get simpler accounting by keeping both ChatGPT Work and Codex on the same plan, and teams standardized on Claude for chat and research get the equivalent benefit from Claude Code.
Both agents shipped real changes in the week before this comparison was written. Claude Code released four versions between November 13 and 19, 2026. Version 2.1.328, released November 19, added an auto mode for Claude Opus 4.6 and Sonnet 4.6 on third-party providers and an onFailure: "block" option for mcp_tool hooks. Version 2.1.327 (November 17) added a confirmation step before /install-github-app changes a repository and let claude mcp add-json read server JSON from standard input. Version 2.1.326 (November 16) added Code Mode support alongside agent teams and tightened Bash and PowerShell permission checks. Version 2.1.325 (November 13) made sandboxed Bash follow the normal permission flow by default and added a prompt before read-only git commands in repositories the user has not trusted.
OpenAI's ChatGPT rate card, updated around November 18, 2026, lists GPT-6.2 Astra and GPT-6.2 Sol as the current models behind ChatGPT Work and Codex tasks, alongside GPT-6.1 Terra and GPT-6.1 Sol, with GPT-5.3-Codex still handling automated code review specifically. The same rate card references a $500 a month ChatGPT Pro tier alongside the existing $200 tier, tied to Ultrafast processing access in Codex.
Codex has the edge here. Its GitHub App auto-reviews pull requests, comments inline on specific lines, and responds when someone tags @Codex on an issue or pull request thread, a workflow Builder.io's Steve Sewell preferred over Claude Code's GitHub integration in his own test of both tools.
Yes, and teams do this often. A common split is Codex for background GitHub tasks like automated pull request review, and Claude Code for terminal sessions where sub-agents, hooks and slash commands matter more.
Since September 19, 2026, Claude Code reads a project's AGENTS.md file in place of CLAUDE.md when no CLAUDE.md exists, according to Anthropic's own documentation. Before that change, Claude Code only read CLAUDE.md.
The Codex CLI and SDK are open source under the Apache-2.0 license, published at github.com/openai/codex with 135.8k stars. Claude Code is not open source. It ships as a closed source npm package, and its GitHub repository holds plugins and documentation rather than the tool's source.
Claude Code is bundled into Claude's Pro plan, which costs $17 a month billed annually or $20 billed monthly, with higher Max ($100) and Ultra ($200) tiers for heavier use. Codex is bundled into ChatGPT plans; OpenAI's own product page lists its Plus tier at €23 a month, with a Pro tier above it.
Developers weighing the claude code vs codex decision can browse both tools, and other coding agents, in the coding category on AI Agents Listing.
One email a week. New agents, MCP servers and skills, and what is actually getting traction.
AI agent vs chatbot compared on control flow, tool use and autonomy, with sourced definitions from Salesforce, Google Cloud and Wikipedia.
6 min read
AI agent examples compared: Claude Code, Devin, ChatGPT agent, Intercom's Fin, Salesforce Agentforce and Manus, each sourced from the vendor's own docs.
8 min read
MCP logging lets a server send log messages to its client. The 2026-07-28 MCP spec deprecated it in favor of stderr logging and OpenTelemetry.
6 min read