Estimated Belliste score, based on the available capabilities, quality, features and pricing information. It is an indicative profile estimate, not an independent benchmark.
How this profile estimate is calculated
Capabilities contribute 35%, quality wording 30%, listed features 20%, and pricing 15% when available. When capability fields are absent, specialization breadth supplies a 25% fallback. Missing categories are omitted and the remaining weights are rescaled. The estimate uses each model’s own published profile; it does not measure answer accuracy or replace a hands-on test.
OverviewOpenAI CodexClaude Code
PriceFree $0/month · Go $8/month · Plus $20/month; monthly USD subscription prices. API usage is billed separately.Pro $20/month billed monthly · $200/year prepaid (about $16.67/month) · API access billed separately by usage.
Free planFree access is included with limited model/client availability; Free and Go currently list GPT-6 Luna Standard in desktop, subject to rollout. A universal paid-plan trial duration is not disclosed.Claude Free does not include Claude Code. No standard Claude Code free-trial duration is disclosed; promotional offers may differ.
Paid plans
Go $8/month · Plus $20/month · Pro $100, $200 or $500/month. Business: $20/user/month billed annually, or $25 monthly, minimum 2 users. Enterprise/Edu: contact sales. Model and interface access vary; API usage is separate.
Pro: $20/month or $200/year prepaid. Max: from $100/month, with 5x or 20x Pro usage options. Team: Standard $25/seat/month ($20 billed annually); Premium $125 ($100 billed annually). Enterprise: sales-assisted pricing and usage terms. Taxes may apply.
Usage limitsAllowance depends on model, reasoning, speed and task. Plus/Standard Business publish 5-hour estimates; weekly limits may apply. Pro currently has no 5-hour limit. Extra credits and API billing are separate.Subscription allowances vary by plan, model and task; usage is shared with Claude. Team/Enterprise seats have rolling 5-hour and weekly allowances. Optional usage credits and API usage are billed separately.
QualityCan inspect changes and run project checks when tools are available. Results depend on model, repository context and verification quality. Belliste has not run a comparative benchmark for this profile.Supports repository work, test execution and review workflows. Output quality varies with the model, project context and task. Belliste has not conducted a comparative coding benchmark; generated changes require project-specific validation.
FeaturesMulti-file code generation and editing · Repository exploration and debugging · Test generation and execution · Local and connected pull-request review · Terminal commands and development tools · Planning, permissions and sandbox controls · Parallel subagents and Git worktrees · Local and cloud execution · AGENTS.md instructions and optional memory · MCP, plugins and external integrations · Screenshot-guided coding · CLI, SDK and app-server automationMulti-file code generation and editing · Bug investigation and refactoring · Test and documentation generation · Terminal commands and automated checks · Plan Mode and configurable approvals · Parallel agents and background sessions · Local and cloud execution · MCP integrations, skills and project rules · Screenshot-to-code workflows · Git and pull-request workflows
Overview & Compatibility
9 specifications +
SpecificationOpenAI CodexClaude Code
Starting PriceFree $0/month · Go $8/month · Plus $20/month; monthly USD subscription prices. API usage is billed separately.Pro $20/month billed monthly · $200/year prepaid (about $16.67/month) · API access billed separately by usage.
Tool TypeCoding Agent / IDE Assistant; local development and cloud task workflows.Coding Agent / IDE Assistant; available through CLI, editor integrations, desktop and web.
Primary Use CaseImplement features, investigate bugs, review changes and automate repository maintenance.Implement features, investigate bugs, refactor repositories and automate development work with an agent that can edit files and run tools.
Best ForIndividual developers and engineering teams working with repositories, tests and code review.Developers and engineering teams working on existing repositories, multi-file changes and repeatable coding workflows.
Access MethodsDesktop app · CLI · IDE extension/integrations · Web/cloud; mobile cloud access depends on plan and client.CLI · VS Code extension · JetBrains plugin · Claude desktop Code tab · Web; mobile access to cloud sessions.
Supported Operating SystemsWindows, macOS and Linux; CLI also runs in WSL. Installation, sandboxing and desktop capabilities differ by OS.CLI: Windows, macOS, Linux and WSL. Desktop availability depends on OS; the Linux desktop app is beta. Native Windows and WSL differ in sandbox support.
Supported IDEs & EditorsVS Code, VS Code Insiders, Cursor and compatible editors; Xcode and JetBrains offer separate integrations.VS Code and compatible editors including Cursor; JetBrains IDEs through the Claude Code plugin. Other editors can run the CLI in an integrated terminal.
Programming LanguagesWorks across programming languages through source files and installed project tools. No exhaustive language-by-language compatibility guarantee is disclosed.Works across source languages using repository files and installed development tools. An exhaustive supported-language list and equal capability across languages are not disclosed.
Framework SupportUses the project’s framework, dependencies and build tools. A complete framework certification/support matrix is not disclosed.Can work with a project’s framework and tooling. No exhaustive official framework compatibility matrix or framework-specific support guarantee is disclosed.
Coding Capabilities
11 specifications +
SpecificationOpenAI CodexClaude Code
Code CompletionTraditional as-you-type inline code autocomplete: Not disclosed. Codex documents prompted coding, editor-context input and diff review.Traditional as-you-type inline code autocomplete: Not disclosed. Documented editor features include prompted code changes and inline diff review.
Code GenerationGenerates code and files from instructions in an existing repository or new project.Generates source code, new files and feature implementations from natural-language instructions.
Inline Code EditingUses selected editor code and open files as context for requested changes; edits can be reviewed beside the source.Selected code and line ranges can be sent from the editor for prompted changes; proposed edits appear in an inline or side-by-side diff for review. This differs from as-you-type autocomplete.
Multi-File EditingCan change multiple files in one repository task and present the resulting diff for review.Coordinates changes across multiple files in a repository; results should be checked against project tests.
Code ExplanationExplains repository structure, functions and unfamiliar implementation details using accessible files.Explains functions, project structure and execution paths using repository context.
RefactoringRefactors existing code using project instructions and can verify the result with available checks.Refactors existing code and updates dependent call sites; can run project checks after changes.
Debugging AssistanceInvestigates error reports, traces code paths and implements fixes; available logs and tests improve verification.Investigates errors, searches relevant code and proposes or implements fixes using logs and project tools.
Test GenerationCan write tests for requested behavior and regressions using the project’s testing tools.
Agents & Execution
9 specifications +
SpecificationOpenAI CodexClaude Code
Agent Mode & PlanningAgent can inspect, plan, edit and verify a task. Planning and execution follow the selected mode and permission policy.Agentic loop gathers context, makes changes and checks results; Plan Mode separates investigation/planning from implementation.
Terminal Command ExecutionRuns shell commands, builds and installed development tools within the allowed filesystem and network scope.Runs shell commands, builds and development tools, subject to permission rules and tools installed in the execution environment.
Automated Test ExecutionRuns available test suites, reads failures and can iterate on fixes; dependencies and test infrastructure must be configured.Runs the project’s installed test tools, reads failures and can iterate on fixes. Test infrastructure and dependencies must be available.
Browser InteractionBuilt-in interactive browser in the desktop app for sites and local previews. It is not built into Codex CLI or IDE extension; browser tools can be added separately where supported.Interactive Chrome integration can navigate pages, enter forms, inspect an app and help debug browser behavior; requires the supported browser setup.
Background / Cloud AgentsCloud tasks run in isolated workspaces and continue while the local computer sleeps. Local background work depends on the host remaining available.Background local work and cloud sessions are supported. Hosted cloud tasks continue independently of the local laptop and share account usage limits.
Parallel Agent TasksSupports parallel subagents and separate concurrent tasks; Git worktrees help isolate repository changes. Parallel work increases usage.Parallel subagents and concurrent sessions are supported; available capacity and consumption depend on the plan and runtime.
Human Approval ControlsPermission profiles control file, network and execution access. Requests outside permitted scope can require approval; managed settings can constrain available policies.
AI Models & Codebase Context
8 specifications +
SpecificationOpenAI CodexClaude Code
Supported AI ModelsGPT-6.1 Sol, GPT-6 Sol, GPT-6 Luna and GPT-6 Astra where eligible. GPT-5.6 variants remain during rollout; model access depends on plan, client and sign-in method.Anthropic API aliases currently resolve to Opus 5.5, Sonnet 5.5 and Haiku 5.5. Fable 5.1 and Fable 5 can be selected where eligible. Provider, plan and CLI version affect access.
Model SelectionConfigurable defaults plus manual model/reasoning selection via the picker, /model or --model. Available settings depend on client and workspace policy.Manual model picker and /model command; configured defaults, fallback chains and opusplan can switch models automatically.
Bring Your Own API KeySupports OpenAI API-key authentication in CLI, SDK and IDE workflows. API billing is separate; ChatGPT-dependent cloud features are limited or unavailable with API-key sign-in.Supports an Anthropic API key and credentials for supported cloud providers. API/provider usage is billed under the corresponding account.
Local Model SupportCLI OSS mode supports local providers such as Ollama and LM Studio via --oss. Hardware, model tool support and feature compatibility vary.Anthropic does not support routing Claude Code to non-Claude models through gateways. Unofficial local-model adapters are outside official support; running the CLI locally does not make Claude inference local.
Codebase Context & IndexingReads and searches accessible repository files, uses editor selections and project guidance. A universal persistent semantic-index specification is not disclosed.Builds repository context through file reads, filename/content search and optional code-intelligence plugins. A separate persistent semantic-index specification is not disclosed.
Context WindowModel/provider dependent; context usage and compaction affect a session. No single fixed Codex-wide context window is disclosed; API model maximums do not guarantee equivalent usable product context.
Developer Workflow & Integrations
5 specifications +
SpecificationOpenAI CodexClaude Code
Git & Pull RequestsLocal Git diff, branch/worktree and change-review workflows. Authenticated Git tools/integrations can prepare and create pull requests; remote-provider permissions apply.Local Git branches, diffs and commits; PR creation with authenticated tooling. Cloud repository cloning and PR creation require GitHub; other remotes have cloud limitations.
MCP & External IntegrationsMCP supports external tools and data sources. Plugins and integrations can connect GitHub, GitLab beta, Slack, Linear and other services; availability and permissions differ.MCP servers extend tool access; integrations include editor plugins, GitHub Actions, GitLab CI and Slack. Third-party accounts and permissions may be required.
Web SearchWeb search is supported, with availability and network behavior determined by client, configuration and workspace policy.Built-in web search and page fetching for documentation, error research and other development context; network and organization policies apply.
API / SDK AccessCodex SDKs for TypeScript and Python, app-server for custom clients, and codex exec for automation/CI. Authentication and usage billing follow the chosen runtime.Programmatic use through the Claude Agent SDK and non-interactive CLI; SDK documentation is provided for TypeScript and Python. API/provider billing is separate from subscription allowances.
Deployment SupportCan run deployment commands or connected tools with permission. Sites adds a build/publish workflow on eligible clients and plans; external hosting has separate credentials and costs.Runs user-configured deployment scripts, provider CLIs or MCP tools with permission; no hosting service is included. Deployment credentials and infrastructure costs depend on the chosen provider.
Pricing, Teams & Security
6 specifications +
SpecificationOpenAI CodexClaude Code
Free Plan / TrialFree access is included with limited model/client availability; Free and Go currently list GPT-6 Luna Standard in desktop, subject to rollout. A universal paid-plan trial duration is not disclosed.Claude Free does not include Claude Code. No standard Claude Code free-trial duration is disclosed; promotional offers may differ.
Usage Limits / Extra UsageAllowance depends on model, reasoning, speed and task. Plus/Standard Business publish 5-hour estimates; weekly limits may apply. Pro currently has no 5-hour limit. Extra credits and API billing are separate.Subscription allowances vary by plan, model and task; usage is shared with Claude. Team/Enterprise seats have rolling 5-hour and weekly allowances. Optional usage credits and API usage are billed separately.
Team AdministrationBusiness/Enterprise provide workspace administration; managed configuration controls Codex access, permissions, models and rollout. Enterprise adds broader governance and audit capabilities.Team/Enterprise support centralized billing, member management and organization settings; Enterprise adds role-based controls and compliance capabilities.
Code Privacy & Data RetentionCloud inference processes supplied context unless a local provider is configured. Business/API data is not used for training by default. Retention varies by service, plan and controls; API abuse-monitoring logs are generally retained up to 30 days, with exceptions and eligible ZDR. No universal Codex retention duration is disclosed.Remote inference processes selected context. Consumer training follows the user setting: retention 30 days when disabled, 5 years when enabled. Commercial data is not used for training by default; standard retention 30 days. Eligible Enterprise organizations can request ZDR; exceptions apply.
Self-HostingLocal Codex CLI/app-server can run on infrastructure you manage, including local-model setups. This does not self-host the managed Codex Cloud service; hosted-service parity is not disclosed.Self-hosted execution environments are public beta for Team/Enterprise and require admin enablement. They do not provide offline Claude model inference; cloud inference and feature restrictions still apply.
Explore more comparisons
Compare each model with other close options in the same AI category.
Creates tests for functions and modules, including edge cases, using the project test framework.
Documentation GenerationCan draft or update README files, code comments and project documentation from repository context.Drafts docstrings, API documentation, README content and other project documentation using the supplied code and instructions.
Code ReviewOn-demand /review for local Git changes; connected PR reviews and automatic GitHub reviews require the relevant integration, permissions and account access. GitLab review integration is beta.On-demand /code-review in a session; hosted GitHub PR Code Review for Team/Enterprise after setup, with separate usage billing. GitHub Actions and GitLab CI workflows are also supported.
Image-to-CodeAccepts screenshots, diagrams and mockups as input for code changes and UI implementation; supports image input in CLI and editor workflows.Accepts screenshots and mockups as context and can generate HTML/CSS or other code matching the supplied design.
Configurable permission modes and allow/deny rules. Manual mode requests file-edit and command approval; Plan mode supports reviewing the approach before implementation.
Execution Isolation / SandboxmacOS uses Seatbelt; Linux/WSL2 use the Linux sandbox; native Windows has a Windows sandbox. Cloud tasks get isolated workspaces. Isolation depends on configuration.Optional shell sandbox isolates filesystem/network access on macOS, Linux and WSL2; native Windows shell execution is unsandboxed. File tools, MCP and hooks are outside that shell boundary. Hosted sessions use isolated VMs.
Execution EnvironmentBoth: local CLI/IDE/desktop execution and Codex Cloud tasks. Remote-model inference is separate from where commands execute.Both: local CLI/IDE/desktop sessions and cloud sessions. Local execution still uses remote model inference under supported deployments.
Model/provider dependent: current Fable, Opus 5.5, Sonnet 5.5 and Haiku 5.5 support 1M tokens. Older variants can use 200K or an eligible [1m] option; plan settings and compaction affect usable context.
Custom Instructions / RulesAGENTS.md provides repository and directory instructions; user/project configuration and execution rules add workflow constraints.CLAUDE.md files and scoped .claude/rules instructions define project conventions; settings can also be applied at user, project and organization scopes.
Persistent Project MemoryOptional local memory carries useful context across eligible chats. Desktop, CLI and IDE use the host’s local store; it is separate from ChatGPT web memory.Auto memory saves useful project information across sessions. Memory is stored separately from the active conversation and user-maintained instruction files.
SSO / Enterprise SecurityBusiness includes SAML SSO and MFA. Enterprise/Edu add controls such as SCIM, role-based access, audit/compliance access and retention/residency settings; exact scope depends on plan.SSO on Team/Enterprise; additional enterprise access and compliance controls. Anthropic lists SOC 2 Type II, ISO 27001:2022 and ISO/IEC 42001:2023; verify service scope and contractual requirements.
