Anthropic Claude Code vs OpenAI Codex (2026)

The GitHub Action runs on any GitHub event or cron schedule without a mention, beside @claude mentions and terminal, desktop, IDE, web and headless runs.

Six invocation surfaces, from the desktop app and CLI to Remote mode on a phone, plus scheduled tasks, a GitHub Action, non-interactive CI mode and third party integrations for GitHub, GitLab, Slack and Linear.

Context is assembled per run from the repository, @ mentioned files, MCP servers and a web search cache over OpenAI's index of the public web, and no maintained retrieval structure over the customer's own code or documents is documented.

Auto memory, written by Claude as it works and scoped per repository, persists across sessions until edited or deleted and is browsable and editable through /memory.

Codex writes its own local memory files from eligible prior chats, with a stated scope under CODEX_HOME and a stated lifetime through memories.max_unused_days, off by default and inspectable as files.

trust.openai.com lists SOC 2 Type 2 and ISO/IEC 27001 and 42001 among other attestations, scoped to the API Platform and ChatGPT Enterprise, Edu and Team rather than Codex by name, with managed configuration, SCIM, EKM and a Compliance API on the access side.

OpenTelemetry exports metrics, events and traces with a decision event for every permission prompt and a result event for every tool call, and cloud sessions log all operations for audit.

Opt in OpenTelemetry export carries a documented event catalog of tool decisions, tool results and approval outcomes to a customer controlled collector, and Enterprise adds a Compliance API with audit events, though telemetry is off by default and prompt text is redacted unless enabled.

Claude Code runs on the developer's own machine by default, cloud sessions run in isolated Anthropic VMs or on the organization's self hosted environments, and data is encrypted in transit but not at rest.

Codex runs locally on the developer's machine under an OS enforced sandbox with dev container support, and Enterprise adds Private Link, IP allowlisting, mutual TLS and Amazon Bedrock as a model provider, while cloud tasks run in OpenAI managed containers rather than the customer's own cloud.

Plugins bundle skills, subagents, hooks and MCP servers and are browsable in Anthropic's official marketplace, beside bundled skills and built in subagents.

Skills and plugins are installable surfaces with build guides, a submission path, enterprise plugin management and skill controls, and Claude Code plugins can be submitted too.

Every model is Anthropic's: customers choose a Claude model and where it is served, but not the model maker.

Local Codex clients accept custom model providers, with Mistral's API and a local Ollama endpoint in the vendor's own example and an OSS mode for Ollama or LM Studio, so the customer chooses across model makers.

claude plugin eval runs prompt suites in fresh isolated sessions, grades them against a no plugin baseline and can gate CI on the score.

Code review, the Codex Security plugin and CI scanning check the customer's code and Auto-review gates individual actions at runtime, but no harness to evaluate, replay or regression test agent behavior is documented.

The Chrome integration lets the agent operate a visible browser session with the user's logins, and computer use extends this to native macOS apps.

Browser and Computer use are documented capabilities with dedicated pages, alongside a browser extension and Appshots, and the security page treats their activity as a distinct traffic surface with its own feature controls.