OpenAI Codex

OpenAI Codex should be presented as a coding-agent product family rather than a single narrow CLI utility. The current data layer, based on readable OpenAI Developers pages and the openai/codex GitHub repository, represents Codex across CLI, IDE extension, web/cloud, desktop/app-like, and cloud-agent workflows. It is a strong first candidate for teams that already use OpenAI infrastructure or want to compare terminal, IDE, and hosted coding-agent surfaces under one product family. The most important copy boundary is open source. Codex is not safely described as fully open-source. The Codex CLI repository is fieldized as Apache-2.0 open source, but hosted/web/cloud Codex services are proprietary/hosted product surfaces. Pricing/free-tier signals exist in the data layer, but prices, credits, limits, account requirements, API-key mode, and models are high-change. Prior review also found openai.com/codex/ and a Help Center CLI article blocked by Cloudflare/403 in the review environment, so the page should rely on Developers/GitHub sources and preserve that source-access caveat.

Official ClaimManually ReviewedLast checked 2026-07-04Expansion reviewed 2026-07-08

Visit official website Compare tools Use selector

Quick verdict

Best fit: teams evaluating OpenAI coding-agent workflows across terminal, IDE, browser/cloud, and app-like surfaces. Main caveat: open-source claims must be scoped to the Codex CLI repository, not the whole product family.

Who should evaluate this tool first

Your team wants Codex CLI plus possible IDE/cloud workflows and can review official OpenAI Developers/GitHub source boundaries.
Your team already uses OpenAI infrastructure or wants to compare terminal, IDE, and hosted coding-agent surfaces under one product family.
You can distinguish ChatGPT sign-in, API key mode, workspace settings, sandbox/approval boundaries, and account-specific usage limits.

Who should not choose it first

You need a fully open-source product.
You need fully local inference.
You need a single-surface CLI-only tool with no hosted component.

Data card

Tool type: Cli, Ide Extension, Web App, Desktop App, Cloud AgentCategories: Code Generation, Repo Edit, Bug Fixing, Test Generation, Workflow AutomationPlatforms: Terminal, Browser, Local Repo, Cloud, Vs Code, Jetbrains, Cursor, Windsurf, Macos, Windows, LinuxPricing: Freemium; free tier YesOpen-source: Partial Open SourceLocal support: PartialModels: OpenAI Models, API Key Dependent

Capabilities

McpSupport: LimitedTerminalExecution: YesRepoEdit: YesTestLoop: LimitedPrIssueSupport: LimitedMultiFileEdit: YesBrowserUse: LimitedLongRunningTasks: Limited

Workflow scenarios

Workflow scenarios worth testing

Use these as evaluation tasks, not benchmark claims.

Cli repo tasks

Use CLI repo tasks as an evaluation task for this tool; treat observed results as workflow fit signals, not benchmark results.

Ide-side coding assistance

Use IDE-side coding assistance as an evaluation task for this tool; treat observed results as workflow fit signals, not benchmark results.

Hosted/cloud task delegation

Use hosted/cloud task delegation as an evaluation task for this tool; treat observed results as workflow fit signals, not benchmark results.

Api-key-mode evaluation

Use API-key-mode evaluation as an evaluation task for this tool; treat observed results as workflow fit signals, not benchmark results.

Openai-vs-anthropic workflow comparison

Use OpenAI-vs-Anthropic workflow comparison as an evaluation task for this tool; treat observed results as workflow fit signals, not benchmark results.

Setup/adoption notes

Distinguish ChatGPT sign-in, API key mode, workspace settings, sandbox/approval boundaries, and account-specific usage limits.
Specify which Codex surface a claim applies to: CLI, IDE extension, web/cloud, desktop/app-like, or cloud-agent workflow.

Existing fit summary

Developers and teams evaluating OpenAI coding-agent workflows across terminal, IDE, and cloud-based software tasks.

Limitations to keep visible

openai.com/codex/ and the referenced Help Center CLI article remained unreadable in this environment on 2026-07-04 due to Cloudflare challenge / HTTP 403; do not claim those exact pages were manually read.
Pricing, models, usage limits, and credit extensions are high-change fields; consult the official Codex pricing/auth pages before purchase or quota decisions.
Codex CLI repository is Apache-2.0 open source, but Codex Web/Cloud and hosted product services are proprietary/hosted; openSourceStatus is therefore partial rather than fully open-source.
Account-specific limits, organization settings, Enterprise contract terms, region availability, and actual workspace admin configuration still require account-level confirmation.

Privacy/local caveats

Local repo access means the tool can work with files in a local workspace. It does not prove local model inference, self-hosting, no data transfer, or no telemetry. Confirm current privacy, data-use, retention, training-use, and enterprise controls from official sources for your plan before using sensitive code.

Codex CLI can operate from a local repo, but hosted/web/cloud services and model processing are not fully local; sandboxing/security details should cite Developers docs.

Local repo access means the tool can work with files in a local workspace. It does not prove local model inference, self-hosting, no data transfer, or no telemetry. Confirm current privacy, data-use, retention, training-use, and enterprise controls from official sources for your plan before using sensitive code.
Enterprise controls, retention/residency, organization settings, and contract terms require official/account-level confirmation.

OpenAI Developers Codex auth/pricing/sandboxing/security documentation was readable on 2026-07-04 and confirms plan-dependent workspace permissions, RBAC, retention/residency, admin/security controls, and API-key-mode data-setting boundaries. The original openai.com/codex marketing page and the Help Center CLI article still returned Cloudflare challenge / HTTP 403 in this review environment, so their page-specific claims are retained only as caveated failed-access evidence.

Pricing caveats

Developers Codex pricing/auth pages were readable on 2026-07-04, but prices, credits, models, plan limits, and API-key mode are high-change.

Pricing, models, usage limits, credits, and account-specific settings are high-change.
Avoid exact prices, credits, limits, and plan claims unless they are freshly source-reviewed.

Pricing signal in the data layer: Freemium; free tier Yes. Exact prices, quotas, plan names, included models, and enterprise terms can change quickly.

Open-source/local boundaries

Codex CLI repository is open-source, but Codex web/cloud/hosted services are proprietary; copy must avoid calling the whole product open-source.

Partially open-source: the current data layer treats Codex as partial-open-source because the Codex CLI repository is Apache-2.0 open source, while hosted/web/cloud Codex services are not fully open-source.
Codex CLI can work from a local repo, but hosted/cloud services and model processing are not fully local.

Open-source status in the data layer: Partial Open Source. Local support signal: Partial.

Related source-supported pages

OpenAI Codex vs Claude Code

ready-for-copy-with-caveat. OpenAI marketing/help pages had Cloudflare challenge in prior review; use Developers/GitHub sources for Codex and state caveat. Claude Code pricing remains needs review.

Best AI coding agents for CLI workflows

ready-for-copy-with-caveat. Use best-fit language by workflow; do not claim measured performance or ranking. Gemini CLI billing/privacy still needs Google docs review.

Open source AI coding agents

ready-for-copy-with-caveat. OpenAI Codex is partial-open-source only; hosted/cloud services are not fully open-source. BYO/local privacy depends on provider configuration.

FAQ

Is OpenAI Codex open-source?

Partially. The current data layer treats Codex as partial-open-source because the Codex CLI repository is Apache-2.0 open source, while hosted/web/cloud Codex services are not fully open-source.

Is Codex only a CLI tool?

No. The current data layer includes CLI, IDE extension, web/cloud, desktop/app-like, and cloud-agent surfaces. Copy should still specify which surface a claim applies to.

Does Codex run fully locally?

Do not assume that. Codex CLI can work from a local repo, but hosted/cloud services and model processing are not fully local.

Can I quote Codex pricing on this page?

Avoid exact prices, credits, limits, and plan claims unless they are freshly source-reviewed. Use freemium/free-tier signals with a high-change caveat.

How should Codex be compared with Claude Code?

Use /compare/openai-codex-vs-claude-code/ and compare by workflow surface, CLI behavior, IDE/cloud needs, account/security controls, and open-source boundaries rather than vendor popularity.

Evidence freshness note

Evidence note: This page uses Agentic Code Hub’s current data layer, source-review notes, and public/official sources where available. Fields marked yes, limited, unknown, ready-for-copy-with-caveat, or needs-source-review are not benchmark results. Confirm official docs, pricing, privacy/security terms, account requirements, and product availability before adopting a tool. Developers/GitHub sources were readable; openai.com/codex/ and the referenced Help Center article remained unreadable in the 2026-07-04 review environment due to Cloudflare/403.