Codex vs Claude Code: which coding agent for your team?
OpenAI Codex and Claude Code compared: plans, prices, admin controls, data and security. Vendor statements only, with date and source.
Sebastian LangOctober 6, 202616 min read
If you have to commit a team to one coding agent, you quickly end up with the question Codex vs Claude Code. Both tools read your project's code, change files, run commands and work in the terminal, in the IDE, in a desktop app and in the cloud. The differences lie less in the interface than in the plans, the admin functions and the data commitments, and that is exactly what we compare here.
Transparency note: we train teams in Claude Code and Cowork. That is why we compare only statements the vendors publish themselves, with date and source. We do not rate model quality, name no winner and cite no benchmarks. The same criteria apply to both products, and where one side does not state something, we say so.
As of 6 October 2026
TL;DR
- Both agents are included in existing plans: Codex in the ChatGPT plans, Claude Code in all paid Claude plans. For both, cloud features are available only with an account sign-in, not with a plain API key.
- Team seat prices are close together on 6 October 2026: ChatGPT Business 20 USD per user per month (annual, from 2 users), Claude Team Standard 20 USD per seat per month (annual). The two vendors measure usage differently, though, so the seat price says little about total cost.
- The decision depends on your platform, your governance requirements and your tasks. A pilot running the same tasks in both tools tells you more than any table.
In brief: what Codex and Claude Code are
OpenAI Codex consists, according to OpenAI's documentation, of the Codex CLI, an IDE extension and cloud automation, plus the ChatGPT desktop app. The CLI is meant to review code, change it, run commands and automate recurring work without leaving the terminal. According to OpenAI, Codex and ChatGPT Work share the same prices, credits and usage limits (OpenAI, Codex Pricing, Codex CLI).
Claude Code is, according to Anthropic, an agentic coding tool that reads the codebase, edits files, runs commands and integrates with development tools. It runs in the terminal, in IDEs, in the desktop app and in the browser (Claude Code Docs, Overview). How the agent works is explained in What is Claude Code?.
Both read project-specific instruction files. Codex supports AGENTS.md; Claude Code reads CLAUDE.md and, according to the documentation, can also read an existing AGENTS.md, on its own or together with CLAUDE.md (OpenAI, feature overview, Claude Code Docs, Memory).
Where the agents work: terminal, IDE, desktop, cloud, GitHub
Both products cover the same places to work, with differences in the details. The table lists what the documentation explicitly names on 6 October 2026. If something is missing, that only means we did not find it on the pages we checked.
| Place to work | OpenAI Codex | Claude Code |
|---|---|---|
| Terminal | Codex CLI, also non-interactive with codex exec | Claude Code CLI, also non-interactive with claude -p |
| IDE | IDE extension (VS Code Marketplace) | VS Code and JetBrains extension |
| Desktop | ChatGPT desktop app (macOS, Windows; Linux according to the guide) | Desktop app with Code tab, includes Claude Code |
| Cloud | Codex cloud on the web and in the desktop app, with its own cloud environments | Cloud sessions in the browser, in the desktop app, on mobile and via claude --cloud |
| GitHub | Code review and tasks via @codex in pull requests | GitHub Actions, also GitLab CI/CD |
| Chat integration | Slack and Linear | Slack (Pro and Max) |
| Mobile | iOS app, remote control of a connected computer | Code tab in the Claude app |
| Plain API key | CLI, SDK and IDE, but no cloud features such as GitHub review or Slack | CLI and local features, but no cloud sessions, Code Review or Routines |
Sources: Codex Pricing and feature matrix, Codex Cloud, Codex Code Review on GitHub, Claude Code Overview, Claude Code Feature Availability, Claude Code on the web.
Two details help with planning. First, both require signing in with an account with the respective vendor to use cloud features. For the API key, Codex explicitly states "No cloud-based features", and Claude Code lists cloud sessions, Code Review, Routines and Desktop under "Features that require a Claude subscription". Second, Claude Code can also be operated through Amazon Bedrock, Google Cloud Agent Platform or Microsoft Foundry (Claude Code, Admin setup). According to OpenAI's documentation, Codex's local interfaces support direct access through Amazon Bedrock (OpenAI, Codex with Amazon Bedrock).
Claude Code
Set up Claude Code properly in your engineering team
Models and plans with prices
Codex lives in ChatGPT plans, Claude Code in Claude plans. The table puts the same rows side by side for both. All prices are list prices in USD as the vendors show them on 6 October 2026. Anthropic quotes prices excluding taxes.
| Row | OpenAI Codex (ChatGPT plans) | Claude Code (Claude plans) |
|---|---|---|
| Entry for individuals | Plus: 20 USD per month | Pro: 17 USD per month with annual payment (200 USD upfront), 20 USD monthly |
| More usage for individuals | Pro: from 100 USD per month (tiers of 100, 200 and 500 USD) | Max: from 100 USD per month (5x or 20x the usage of Pro) |
| Team | Business: 20 USD per user per month with annual billing (from 2 users), 25 USD with monthly billing | Team Standard seat: 20 USD per seat per month annually, 25 USD monthly (teams of 2 to 150); Premium seat: 100 USD per seat per month with annual billing, 125 USD with monthly billing |
| Enterprise | Enterprise and Edu: price on request (sales) | Enterprise: 20 USD per seat per month, annually, plus usage at API prices |
| Usage via API key | Billed at API prices, without cloud features | Billed via Console at API prices |
| How usage is measured | Allowance included in the plan (estimate in messages per 5 hours, weekly limits possible); beyond that ChatGPT credits; Enterprise with flexible pricing without fixed limits, usage via credits | Allowance included in the plan in a rolling 5-hour window plus weekly limits, chat and Claude Code draw from one pool; beyond that usage credit at API prices; Enterprise: all usage at API prices |
Sources: OpenAI, Codex Pricing, Anthropic, Claude Pricing, Claude Code Docs, Costs. For more on the Claude pricing model, see What does AI cost per team in 2026?.
OpenAI itself points out that API token prices are separate from usage under a subscription and are not suitable for estimating the tasks a plan includes. The following table shows API list prices and is therefore relevant only where billing is at API prices: with the API key, with Claude usage credit and with Claude Enterprise. OpenAI bills ChatGPT credits for Codex according to its own rate card, which is on the pricing page and is not to be equated with this table.
| Model | Input per million tokens | Output per million tokens |
|---|---|---|
| GPT-6 Astra | 10 USD | 50 USD |
| GPT-6.1 Sol | 2 USD | 10 USD |
| GPT-6 Luna | 0.10 USD | 0.50 USD |
| Claude Fable 5.1 | 10 USD | 50 USD |
| Claude Opus 5.5 | 4 USD | 20 USD |
| Claude Sonnet 5.5 | 2 USD | 10 USD |
| Claude Haiku 4.5 | 1 USD | 5 USD |
As of 6 October 2026, list prices according to the vendors' model pages. The OpenAI prices are standard prices for requests of up to 272K input tokens; for GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, requests with more than 272K input tokens cost, according to OpenAI, 2 times as much for input and 1.5 times as much for output. The table says nothing about quality or consumption per task. Sources: GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna, Claude Models Overview. A broader overview is given by the LLM comparison 2026.
Which models do the agents use?
According to OpenAI, Codex recommends GPT-6.1 Sol for complex coding and agent workflows, where it is available for the account and client, and GPT-6 Luna for targeted, repeatable tasks. GPT-6 Astra and GPT-6 Sol are also listed, and the GPT-5.6 family remains available during the rollout. GPT-5.5 will be retired on 14 October 2026 in ChatGPT, ChatGPT Work and Codex with ChatGPT sign-in; the OpenAI API is not affected. Anyone who has hard-coded model names in configurations, scripts or scheduled tasks should replace them beforehand (OpenAI, Codex Models).
For Claude Code, the documentation for the Anthropic API shows the aliases opus (Opus 5.5) and sonnet (Sonnet 5.5). Fable 5.1 is not the default model on any plan and must be selected explicitly (/model fable). On Amazon Bedrock and Google Cloud, sonnet currently resolves to Sonnet 4.5, and on Microsoft Foundry opus resolves to Opus 4.6 (Claude Code Docs, Model configuration). Model lifecycles apply here too: according to Anthropic, Claude Haiku 4.5 will be retired "not sooner than October 15, 2026" (Claude Models Overview).
Admin controls: policies, models, permissions
Both vendors separate access (who may use the tool) from runtime permissions (what the agent may do). The table shows what the documentation names for each row.
| Criterion | OpenAI Codex | Claude Code |
|---|---|---|
| Single sign-on | SAML SSO and MFA from Business | Single sign-on in the Team plan list |
| Central policies | requirements.toml as a file; cloud-managed policies in the admin console (Agent Security) from Business | Managed settings: server-side via the admin console (Team and Enterprise) or via MDM, registry and file |
| What can be set | Approval policy, sandbox mode, permission profiles, web search, hooks, allowed MCP servers, plugin sources | Permission rules, sandbox, allowed MCP servers, plugin sources, hooks, sign-in method, minimum version |
| Restricting models | In Enterprise workspaces, GPT-6.1 Sol, GPT-6 Sol and Luna as well as Astra are initially off until an admin enables them | availableModels in the managed settings, plus enforceAvailableModels: true for the "Default" option (note the documented exceptions); in Enterprise additionally organisation-level model restrictions, enforced server-side |
| Roles, SCIM, audit | Enterprise: RBAC, SCIM, EKM, domain verification, audit logs and usage monitoring via Compliance API | Enterprise: role-based access, SCIM, audit logs, Compliance API, custom retention periods, IP allowlisting |
| Spend control | User limits and spend controls for workspaces with credits | Enterprise: admins set spend limits for users and the organisation |
Sources: Codex Pricing and feature matrix, OpenAI, Managed configuration, OpenAI, Workspace model availability, OpenAI, Usage limits and spend controls, Claude Code, Admin setup, Claude Pricing.
Important in practice: for Codex, OpenAI describes model access and runtime permissions as separate layers. Selecting a model in the local configuration grants no access, and model access cannot loosen the sandbox and approval rules. For Claude Code, managed settings take precedence over user and project settings, apart from a few security-relevant exceptions; arrays such as permissions.allow and permissions.deny merge all sources, so developers can extend lists but not shrink them. Both are functions you should test before the rollout, ideally with one allowed and one blocked action per environment. How to set up policies for agents in the company is shown in our AI policy template.
Data and security: what the vendors document
The same criteria apply to both sides here as well. We only reproduce what the vendors document. This does not replace an assessment of whether it is sufficient for your processing.
| Criterion | OpenAI Codex | Claude Code |
|---|---|---|
| Training on business data | Business, Enterprise, Edu and API: according to OpenAI, no training on business data by default. The feature matrix does not list Plus and Pro as including this assurance | Team, Enterprise and API (commercial terms): according to Anthropic, no training on code or prompts unless the customer actively opts in (for example the Development Partner Program). Pro and Max: users decide via a setting |
| Retention | Enterprise: controls for retention and data residency. For ChatGPT Work (according to OpenAI with the same isolation and permission mechanisms as Codex): conversations follow the workspace settings, deleted chats are generally scheduled for permanent deletion within 30 days, with exceptions | Team, Enterprise and API: 30 days retention by default. Local session logs are stored in plain text for 30 days by default under ~/.claude/projects/ (adjustable via cleanupPeriodDays); sessions started or last resumed in Claude Desktop or Cowork are exempt from this limit by default |
| Zero data retention | OpenAI describes ZDR as an API-specific control that does not define retention rules for ChatGPT Work; according to OpenAI, strict ZDR is unavailable for local computer access through Work Cloud | For qualified Enterprise accounts, not included in the standard plan, to be enabled per organisation by Anthropic |
| Data residency and hosting | Enterprise: residency controls, only for eligible content and supported workloads | For the Claude API, Anthropic names "global" (default) or "us" (1.1x price factor) as inference geography, and currently only "us" as workspace geography. On Amazon Bedrock, Google Cloud Agent Platform and Microsoft Foundry, the rules of the respective platform apply |
| Sandbox and network | By default the agent runs locally in an OS-enforced sandbox, network access is off | Sandboxed Bash tool with file system and network isolation, configurable; in manual mode Claude Code asks before changes, in auto mode a classifier model checks actions |
| Cloud execution | Cloud tasks run in configurable cloud environments with controllable internet access | Cloud sessions run by default in VMs managed by Anthropic; outbound traffic there goes through a security proxy. Sessions in a self-hosted environment run on your own infrastructure, where your own network boundaries apply |
Sources: Codex Pricing and feature matrix, OpenAI, ChatGPT Work Cloud security, OpenAI, ChatGPT Work Overview, OpenAI, Agent approvals and security, OpenAI, Codex Cloud, Claude Code, Data usage, Claude Code, Zero data retention, Claude, Data residency, Claude, Pricing, Claude Code, Security, Claude Code on the web, Claude Code, Cloud environments, Claude Code, Sandboxing.
On country availability: OpenAI lists Germany in its list of supported countries for the API, and Anthropic in its list of supported regions. Check with OpenAI before buying whether a specific ChatGPT plan is available in your country (OpenAI, Supported countries, Anthropic, Supported regions).
A note on context: both vendors point to their own sources for security evidence, Anthropic to the Trust Center with a SOC 2 Type 2 report and ISO 27001 certificate, OpenAI to a Codex security whitepaper. Which evidence is enough for your company is something you decide with your data protection and security function. A catalogue of questions for this is provided by 5 security questions for coding agent vendors, and the comparison with further tools by Cursor, Copilot and Claude Code.
How to base your decision
No winner can be derived from the data, and we do not want to construct one. The following criteria help you make the decision based on your situation.
| Criterion | If this applies to you | What it means for the choice |
|---|---|---|
| Platform | You use ChatGPT Business or Enterprise in the company | According to OpenAI, Codex is included in the ChatGPT plans and shares usage and credits with ChatGPT Work (OpenAI, Pricing) |
| Platform | You use Claude Team or Enterprise | According to Anthropic, Claude Code is included in all paid plans and draws from the same allowance as chat |
| Type of task | You want to hand off tasks to run in the background and review pull requests | Both offer cloud execution and GitHub integration. Check in the pilot which tasks run cleanly for you |
| Type of task | You work interactively in the terminal or in the IDE | Both offer permission controls; Claude Code's Bash sandbox is off by default, must be enabled and is not available on native Windows. Your team's habits and skills are decisive |
| Governance | You need SCIM, audit logs, retention rules | Both list these functions under Enterprise. Match the tables above against your requirements |
| Governance | You want to limit models centrally | Both have model enablement for admins, with different mechanisms (workspace switch for Codex, availableModels with enforceAvailableModels and organisation rules for Claude Code) |
| Hosting | You want to work through a cloud platform or a gateway | Claude Code names Bedrock, Google Cloud Agent Platform, Foundry and its own gateway; Codex names Bedrock and gateways. Feature scope and models differ by route |
| Cost pattern | You want to budget fixed seat costs plus usage | Compare the real consumption of your tasks in a pilot instead of list prices |
If your team still needs to become comfortable working with agents, a structured introduction helps: our Claude Code course comprises 22 lessons in 7 modules, and there is a separate course on Cowork. You do not need it to choose between the two tools.
As an approach we suggest: pick five to ten real tasks from your backlog, run them in both tools with the same instructions, record consumption, review effort and result, and then decide. How to measure productivity properly is explained in Measuring AI productivity honestly.
FAQ
Which is better, Codex or Claude Code?
That cannot be answered from the vendors' statements, and we name no winner. Both cover terminal, IDE, desktop and cloud and offer admin functions in the Team and Enterprise plans. The choice depends on whether you already use ChatGPT or Claude in the company, which governance functions you need and how your tasks behave in a pilot.
What does OpenAI Codex cost?
Codex is included in the ChatGPT plans. According to the OpenAI pricing page, list prices on 6 October 2026 were 0 USD (Free), 8 USD (Go), 20 USD (Plus) and from 100 USD (Pro) per month, and for ChatGPT Business 20 USD per user per month with annual billing (25 USD monthly). Enterprise and Edu are available on request. With an API key, OpenAI bills at API prices, then without cloud features. According to the pricing page, Free and Go name only GPT-6 Luna in the desktop app, depending on the rollout.
Is Claude Code included in the Team plan?
Yes. According to Anthropic, Claude Code is included in all paid plans, and the Team plan list names "Claude Code and Claude Cowork". A Standard seat cost 20 USD per seat per month with annual billing (25 USD monthly) on 6 October 2026, a Premium seat with 5x usage 100 USD (125 USD monthly), for teams of 2 to 150 people. Chat and Claude Code draw from a shared usage allowance.
Can you use Codex and Claude Code in parallel?
Technically yes, they are separate tools. Both support AGENTS.md; for Claude Code, loading behaviour and loading together with CLAUDE.md depend on version and configuration. Both support MCP servers, so parts of the configuration can be shared. Expect double the effort, though: two contracts, two admin consoles, two data flows and two cost centres. So set the same policy for both before you approve them for use.
Which coding agent is GDPR-compliant?
We cannot answer this for a product, because compliance depends on your processing. What is documented are controls: no training on business data by default in the business plans of both vendors, retention periods, optional zero data retention for qualified Enterprise accounts at Claude, and residency controls in Enterprise at OpenAI. Whether that is enough for you is something you assess yourselves with your data protection function, for the specific data the agent sees.
Which models does Codex use?
According to OpenAI, Codex recommends GPT-6.1 Sol for complex tasks and GPT-6 Luna for targeted, repeatable work. GPT-6 Astra, GPT-6 Sol and the GPT-5.6 family are also listed. GPT-5.5 will be retired on 14 October 2026 in ChatGPT, ChatGPT Work and Codex with ChatGPT sign-in; the OpenAI API is not affected. In Enterprise workspaces an admin must first enable new models, and with API-key sign-in the selection follows the API models of the key.
Sources
All sources were retrieved on 6 October 2026.
- OpenAI, Codex Pricing and feature matrix: developers.openai.com/codex/pricing
- OpenAI, Codex Models: developers.openai.com/codex/models.md
- OpenAI, Codex Cloud: developers.openai.com/codex/cloud.md
- OpenAI, Codex CLI: learn.chatgpt.com/docs/codex/cli.md
- OpenAI, Pricing (ChatGPT Work and Codex): learn.chatgpt.com/docs/pricing.md
- OpenAI, Code Review on GitHub: learn.chatgpt.com/docs/third-party/github.md
- OpenAI, Managed configuration: learn.chatgpt.com/docs/enterprise/managed-configuration.md
- OpenAI, Workspace model availability: learn.chatgpt.com/docs/enterprise/workspace-model-availability.md
- OpenAI, ChatGPT usage limits and spend controls: learn.chatgpt.com/docs/enterprise/usage-limits.md
- OpenAI, ChatGPT Work Cloud security: learn.chatgpt.com/docs/enterprise/chatgpt-work-cloud-security.md
- OpenAI, ChatGPT Work Overview: learn.chatgpt.com/docs/enterprise/chatgpt-work-overview.md
- OpenAI, Agent approvals and security: learn.chatgpt.com/docs/agent-approvals-security.md
- OpenAI, Codex with Amazon Bedrock: learn.chatgpt.com/docs/amazon-bedrock.md
- OpenAI API, GPT-6 Astra: developers.openai.com/api/docs/models/gpt-6-astra
- OpenAI API, GPT-6.1 Sol: developers.openai.com/api/docs/models/gpt-6.1-sol
- OpenAI API, GPT-6 Luna: developers.openai.com/api/docs/models/gpt-6-luna
- OpenAI API, Supported countries: developers.openai.com/api/docs/supported-countries.md
- Anthropic, Claude Pricing: claude.com/pricing
- Anthropic, Claude Models Overview: platform.claude.com/docs/en/about-claude/models/overview
- Anthropic, Supported regions: anthropic.com/supported-countries
- Anthropic, Data residency: platform.claude.com/docs/en/manage-claude/data-residency
- Anthropic, Pricing (Data residency pricing): platform.claude.com/docs/en/about-claude/pricing
- Claude Code Docs, Overview: code.claude.com/docs/en/overview
- Claude Code Docs, Costs: code.claude.com/docs/en/costs
- Claude Code Docs, Admin setup: code.claude.com/docs/en/admin-setup
- Claude Code Docs, Model configuration: code.claude.com/docs/en/model-config
- Claude Code Docs, Feature availability: code.claude.com/docs/en/feature-availability
- Claude Code Docs, Claude Code on the web: code.claude.com/docs/en/claude-code-on-the-web
- Claude Code Docs, Memory: code.claude.com/docs/en/memory
- Claude Code Docs, Data usage: code.claude.com/docs/en/data-usage
- Claude Code Docs, Zero data retention: code.claude.com/docs/en/zero-data-retention
- Claude Code Docs, Security: code.claude.com/docs/en/security
- Claude Code Docs, Cloud environments: code.claude.com/docs/en/cloud-environments
- Claude Code Docs, Sandboxing: code.claude.com/docs/en/sandboxing
- OpenAI, Models: learn.chatgpt.com/docs/models
Claude Code
Set up Claude Code properly in your engineering team
The Claude Code training takes your team in 22 lessons from tooling setup to multi-agent orchestration. It is part of the licence.
About the author
Co-Founder · Business & Content Lead
Co-Founder of Sentient Dynamics. 15+ years of business strategy (incl. SAP), MBA. Writes about EU AI Act compliance, ROI measurement and how Mittelstand CTOs actually adopt agentic AI.
