The OpenAI field guide
A guide to OpenAI’s products, platforms and tools. See what each name means, where it fits, and which offerings have changed or retired.
51 directory entries across products, tools, model families and features.
How to read this directory ↓Try a broader name or choose All categories. This directory groups model versions and plan tiers into families.
Directory section
The experiences people use directly.
The conversational product for questions, writing, learning, research and working with information.
Personal plans include Free, Go, Plus and Pro. Plans are access tiers, not different products; features and limits vary.
Source reviewed UTC · Not independently tested
A mode inside ChatGPT for carrying larger tasks through to reviewable results using context, files and tools.
Availability and execution differ between web, desktop and managed workspaces. Review permissions before granting access.
Source reviewed UTC · Not independently tested
OpenAI’s coding agent for working on software, reviewing changes and carrying out development tasks.
Available through the ChatGPT desktop app and developer surfaces including CLI, IDE extension and cloud. Access is surface- and plan-dependent.
Source reviewed UTC · Not independently tested
A dedicated space in ChatGPT for health and wellness information and optional connected health data.
The official announcement includes a July 2026 U.S. adult web/iOS rollout update. It is not intended for diagnosis or treatment; check current eligibility.
Source reviewed UTC · Not independently tested
Create, refine and host websites, web apps and games through ChatGPT.
The documentation lists eligible paid plans and beta usage limits. A created draft and a published site are different states.
Source reviewed UTC · Not independently tested
Directory section
Organization offerings and specialist workspaces—not separate foundation models.
A ChatGPT workspace offering for organizations and their teams.
Use the official plan comparison for current seats, billing, included tools and administration. API access and billing are separate considerations.
Source reviewed UTC · Not independently tested
An organizational ChatGPT offering with enterprise deployment and administration workflows.
Model, app and tool access can depend on workspace policy and rollout. Contract-specific terms should be confirmed with OpenAI.
Source reviewed UTC · Not independently tested
The ChatGPT offering for higher-education institutions.
Institutional eligibility, deployment and workspace controls apply. It is distinct from the U.S. K–12 Teachers offering.
Source reviewed UTC · Not independently tested
A workspace offering for verified U.S. K–12 educators and school staff.
Eligibility verification is required. Consult the official source for the current free-access period and school controls.
Source reviewed UTC · Not independently tested
An offering announced for U.S. government agencies, including deployment in their own Azure environments.
The cited launch explains the deployment model; it does not establish current procurement eligibility or accreditation for every deployment.
Source reviewed UTC · Not independently tested
An organizational ChatGPT offering for healthcare teams working on clinical and administrative information.
Separate from consumer ChatGPT Health. Organization configuration, eligible functionality and applicable agreements require review.
Source reviewed UTC · Not independently tested
A ChatGPT offering for eligible individual clinicians, distinct from institution-wide Healthcare deployments.
Check professional eligibility and the scope of supported functionality with OpenAI. This directory is not clinical guidance.
Source reviewed UTC · Not independently tested
An offering focused on financial-services work, combining analysis workflows with financial data and organizational context.
Announced September 10, 2026. Data-provider entitlements and organizational access must be checked separately; this is not investment advice.
Source reviewed UTC · Not independently tested
A platform for operating AI coworkers with shared business context and enterprise governance.
The Enterprise Frontier Program adds deployment support. It is a platform and services offering, not the generic term “frontier model.”
Source reviewed UTC · Not independently tested
Directory section
Specialist products with their own access and review requirements.
A LaTeX-native workspace introduced for scientific writing and collaboration with AI.
The January 2026 announcement describes the product. Follow the source for the product destination; do not assume its original model or limits remain unchanged.
Source reviewed UTC · Not independently tested
A scientific workspace within the Rosalind life-sciences offering, bringing tools and research workflows together.
Workbench is a research preview. GPT-Rosalind model/API access has separate eligibility and organization approval requirements.
Source reviewed UTC · Not independently tested
An application-security agent for finding, checking and helping remediate vulnerabilities.
Documented surfaces include the plugin, CLI, TypeScript SDK and cloud. Review findings and use only authorized repositories and systems.
Source reviewed UTC · Not independently tested
A cybersecurity offering with Daybreak Blue and separately approved Daybreak Red access.
Access is tied to approved identities, organizations, models and surfaces. It is intended for authorized defensive security work.
Source reviewed UTC · Not independently tested
Directory section
APIs, SDKs and integrations for building with OpenAI.
The platform for integrating OpenAI models and tools into your own applications.
Includes APIs, developer resources and account/project controls. A ChatGPT subscription does not define API model access or API charges.
Source reviewed UTC · Not independently tested
Official libraries for calling the OpenAI API from application code in supported programming languages.
These clients make API requests; they are not the agent-orchestration runtime provided by Agents SDK. API authentication, model access and usage charges still apply.
Source reviewed UTC · Not independently tested
Call the OpenAI API from a terminal for repeatable requests, file processing and generated outputs.
The openai command is distinct from Codex CLI: it sends API requests rather than providing the interactive Codex coding-agent experience. Review authentication and API usage before running commands.
Source reviewed UTC · Not independently tested
Use Codex SDK to automate coding jobs, or App Server to build a client around Codex conversations, approvals and streamed events.
These integrate Codex itself, unlike the general API client libraries or Agents SDK. Language support, transport maturity, authentication and sandbox settings differ; review the appropriate guide before deployment.
Source reviewed UTC · Not independently tested
An API for model responses and tool-enabled application workflows.
Choose models and supported tools for the specific request. Your application still owns its permissions and external actions.
Source reviewed UTC · Not independently tested
A managed API for durable agent sessions with an OpenAI-hosted agent runtime.
Environment, sandbox and tool choices are part of the application configuration. It is distinct from the open-source Agents SDK.
Source reviewed UTC · Not independently tested
A developer toolkit for defining agents and orchestrating application workflows.
Your integration controls execution, tools and deployment. See the SDK documentation for supported languages and components.
Source reviewed UTC · Not independently tested
A toolkit for adding a chat interface and interactive widgets to an application.
ChatKit remains available independently of Agent Builder’s deprecation. Backend integration choices still matter.
Source reviewed UTC · Not independently tested
Build extensions using skills, MCP tools and optional user interfaces for ChatGPT and Codex.
Third-party plugins are not OpenAI products. Installation, connected-account permissions and directory review are separate steps.
Source reviewed UTC · Not independently tested
Merchant integrations for product feeds, checkout and delegated payment flows in ChatGPT.
Merchant and payment-provider onboarding requirements apply. Listing a product does not imply checkout is available everywhere.
Source reviewed UTC · Not independently tested
Advertiser tools for campaigns, creatives, product feeds and conversion measurement.
Consult advertiser documentation for access and supported workflows. This directory does not run ads or collect advertising identifiers.
Source reviewed UTC · Not independently tested
Directory section
The underlying models—not additional apps or subscription plans.
The model catalog includes GPT-6 Astra, GPT-5.6 variants and earlier model generations for language, reasoning and coding.
Model versions, API availability and product selectors differ. Use the live catalog for individual models, snapshots and lifecycle notices.
Source reviewed UTC · Not independently tested
OpenAI’s image-generation and editing model family, separate from the ChatGPT interface that can use image tools.
The current catalog lists GPT-Image-2.5 and GPT-Image-2. Older image model availability must be checked against deprecation notices.
Source reviewed UTC · Not independently tested
Models and APIs for conversational voice, audio, transcription, translation and speech generation.
GPT-Live, Realtime, transcription and text-to-speech are different interfaces or tasks. Whisper is a model name, not a separate consumer app.
Source reviewed UTC · Not independently tested
Open-weight models, including gpt-oss-120b and gpt-oss-20b, for running supported workloads in your own environment.
Deployment, hardware, safeguards and license compliance are your responsibility. Open weights are distinct from hosted ChatGPT access.
Source reviewed UTC · Not independently tested
Model families for representing text as vectors and classifying potentially harmful content.
These are building blocks for applications rather than standalone end-user products. Select the specific model and task in the API catalog.
Source reviewed UTC · Not independently tested
Directory section
Find familiar capabilities without mistaking them for standalone products.
Search and longer-form research workflows help gather and synthesize information within supported products.
Check sources and citations yourself. The capability, its user interface and an API model with a similar name are different things.
Source reviewed UTC · Not independently tested
Talk with ChatGPT and, on supported surfaces, start or coordinate work through Voice in Chat, Work and Codex.
Voice is a user-facing experience, not the same product as an audio API. Desktop plan access, rollout and workspace settings apply; paired iOS Remote access requires a desktop host. Review microphone and screen-context permissions.
Source reviewed UTC · Not independently tested
A ChatGPT mode for working through concepts, practicing questions and checking understanding step by step.
The official guide lists web, iOS and Android across ChatGPT plans. Study mode uses the normal plan and model limits and does not apply inside GPTs, Projects or Temporary Chats. Check answers and follow your institution’s rules.
Source reviewed UTC · Not independently tested
Transcribe and summarize meetings, brainstorms and voice notes into reusable notes in ChatGPT.
The official guide currently lists the macOS desktop app with Plus, Pro, Business, Enterprise and Edu. Obtain appropriate consent before recording others and check transcripts and summaries for errors. Record is distinct from a live Voice conversation.
Source reviewed UTC · Not independently tested
Bring browser tabs into ChatGPT and let supported desktop workflows read or act on websites in your signed-in browser.
The guide lists Chrome, Edge, Brave, Opera and Vivaldi. Side chat is not supported in Opera; rollout and workspace controls apply. Review site permissions and sensitive account access. This extension is distinct from the built-in browser and retired Atlas.
Source reviewed UTC · Not independently tested
Build agents for repeatable work, connect tools, test them and share them with teammates in ChatGPT or Slack.
Eligible workspaces can schedule runs or trigger published agents through the Workspace Agents API. Workspace roles, admin settings and connected-app permissions control access. Workspace Agents are distinct from the developer Agents API and the retired ChatGPT agent mode.
Source reviewed UTC · Not independently tested
Customized versions of ChatGPT and the mechanisms for sharing and discovering them.
A third-party GPT is not a separately operated OpenAI product. Workspace policies can limit sharing and connected actions.
Source reviewed UTC · Not independently tested
Features for organizing related work and carrying selected context across conversations.
Context and memory behavior depends on the product and settings. Check the relevant controls before adding sensitive information.
Source reviewed UTC · Not independently tested
Document, code and visual workflows inside ChatGPT, including file outputs and collaborative editing surfaces.
Generating an output does not prove its accuracy or formatting. The separate file-capability record distinguishes documented behavior from tests.
Source reviewed UTC · Not independently tested
Run supported work on a schedule or in response to supported connected-app events.
Connections, permissions, plan access and supported triggers matter. A scheduled task is not evidence that every action can run unattended.
Source reviewed UTC · Not independently tested
A ChatGPT agent announced for working with approved company data sources and business context.
Announced September 10, 2026. Verify organization eligibility and data-source permissions before relying on it.
Source reviewed UTC · Not independently tested
Historical context
Older names remain findable, with lifecycle boundaries made explicit.
The Sora web and app experiences were discontinued on April 26, 2026. The API has a separate September 24, 2026 shutdown date.
These dates describe different surfaces. Consult the official discontinuation notice for export and transition instructions; this is not a current app recommendation.
Source reviewed UTC · Not independently tested
The browser introduced with ChatGPT built in. Its official announcement now explicitly labels Atlas deprecated.
OpenAI directs readers to ChatGPT Work and current release notes. The original launch’s platform and availability claims are historical.
Source reviewed UTC · Not independently tested
An earlier browser-task agent integrated into ChatGPT agent in July 2025. The later ChatGPT agent mode is also no longer available.
These are separate transitions: Operator moved into ChatGPT agent, and the current agent notice points readers to ChatGPT Work. Historical launch copy does not establish current access.
Source reviewed UTC · Not independently tested
The earlier ChatGPT agent mode for multi-step online tasks has been retired. OpenAI now directs users to ChatGPT Work.
The official overview contains the current retirement notice, although older plan and setup sections remain below it. Do not treat those historical limits as a current entitlement. ChatGPT Work and Workspace Agents are listed separately.
Source reviewed UTC · Not independently tested
The earlier DALL·E image-generation family remains relevant to product history and older integrations.
Check model-specific shutdown notices. Current image-family coverage lives under GPT Image; legacy names are not interchangeable model identifiers.
Source reviewed UTC · Not independently tested
Older developer surfaces include Agent Builder, the Assistants API, the Evals platform and self-serve fine-tuning workflows.
Lifecycle dates differ by product and model. Read the deprecation record before starting or migrating an integration; ChatKit is listed separately.
Source reviewed UTC · Not independently tested
Scope & evidence
This edition covers the public product families and named offerings identified in OpenAI’s product pages and developer documentation. It groups individual model snapshots, subscription variants and minor API endpoints; it is not an exhaustive list of every SKU, experiment or third-party integration.
Documented does not mean available to everyone. Status labels summarize the linked sources as reviewed, not a live account check. Announcements can retain historical launch copy; dated updates and retirement notices take precedence. Prices, eligibility and access should be checked at the official destination.
Each entry has one permanent anchor. This directory explains what an offering is; our history covers how OpenAI developed, and the file-capability record examines a particular workflow. Future detailed guides should answer distinct questions.
By OpenAI Updates Editorial Desk · Source review: OpenAI Updates Verification Desk. Content is produced with AI assistance under role-based aliases, not human credentials. Sources were checked; products were not independently tested for this directory. This page is a dated review edition, not a live status service.