# OpenAI Glossary

A–Z glossary of every model, product, API surface, tool, control and concept used across the six OpenAI preparation tracks, tagged by track.

Every term below appears in at least one of the six OpenAI preparation tracks. Each entry gives the definition, why it matters for the assessments, and which tracks lean on it most. This is independent study material built from publicly available OpenAI learning objectives — see the [credential landscape](/openai/credentials/), the [learning sequence](/openai/learning-path/) and the [assessment model](/openai/assessment-model/) for how the Academy assessments and our independent mock exams relate.

Track tags: **F** = AI Foundations · **AP** = Applied AI · **AG** = Agents and Workflows · **API** = API Developer Path · **CX** = Codex · **L** = AI Leadership.

---

## A

**Acceptance criteria** – The specific, testable conditions an output must meet to be accepted. Distinct from a vague goal; they are what a review gate checks and what a workflow contract records. *AP AG CX*

**Accredible** – The badge platform that issues and hosts OpenAI Academy badges once a course assessment is passed at ≥ 80%. Support for badge issues routes to Accredible; course, enrolment and progress issues route to Gradual. On the assessments, know that a badge is a completion credential, not a certification. *F AP AG API CX L*

**Admin API** – OpenAI administrative surface for managing an organisation programmatically: members, projects, keys, and related governance. Distinct from the Analytics and Compliance APIs. Appears in production-operations and governance items. *API*

**Agent (Agents API concept)** – In the Agents API, an Agent is the configured unit: a model, instructions, tools and MCP servers. It is combined with an Environment and run as a Session. Do not confuse the *Agent* config object with the running *Session* instance. *API AG*

**Agent Builder** – A build surface now grouped under the "Legacy APIs" section of the docs. Refer to it as legacy rather than the newest way to compose agents; prefer the Responses API, the Agents SDK or the Agents API. *API*

**Agents and Workflows** – The third Academy course in the Foundations pathway (75–90 min, ChatGPT Work), covering delegation of structured tasks to agents with human oversight. It mirrors the **AG** preparation track. *AG*

**Agents API** – OpenAI-managed agent runtime (`/v1/agents/sessions`, header `OpenAI-Beta: agents=v1`, beta) that runs the Codex harness: OpenAI runs sessions, orchestration, context compaction and recovery; you supply tools and pick the environment. Choose it for durable cloud agents, hosted sandboxes and long-running work. Constraint: US data residency only, no ZDR. *API AG*

**Agents SDK** – Open-source Python/TS framework you run yourself: agent definitions, models and providers, running and sandboxing agents, orchestration, guardrails, results and state, integrations, observability and agent evals. Choose it when you want code-first control and self-hosted execution. Contrast the OpenAI-managed Agents API. *API AG*

**AGENTS.md** – The Codex agent-configuration file placed in a repository that tells the agent how to work in that codebase: conventions, commands, constraints. The single most exam-relevant Codex file for steering behaviour without touching global config. *CX*

**AI Foundations** – The first Academy course (60–75 min, ChatGPT): AI basics, prompting, context and responsible use. Delivered inside ChatGPT as the pilot certification. Mirrors the **F** track. *F*

**AI for College Students** – Standalone Academy course (45 min, ChatGPT) on study planning, group work, writing and career prep. Not part of a pathway. *F*

**AI for Educators** – Standalone Academy course (45 min, ChatGPT) on lesson planning, activities, assessments and communications. Not part of a pathway; distinct from ChatGPT Foundations for Teachers on Coursera. *F*

**AI Leadership** – Standalone Academy course (180 min) on business priorities, governance, roadmaps and adoption, with no single product. Mirrors the **L** track. *L*

**Amazon Bedrock** – A route to run OpenAI models through AWS infrastructure, relevant to enterprises with AWS data-residency or procurement constraints. Appears in Codex administration and API infra items. *API CX*

**Analytics API** – Programmatic access to workspace usage and analytics data, used for governed reporting on adoption and consumption. Distinct from the Admin and Compliance APIs. *CX L API*

**Apply patch** – A Codex/agent tool that edits files by applying a structured patch rather than free-form rewriting, making changes reviewable and reversible. A discriminator on Codex items about safe, auditable edits. *CX API*

**Applied AI Foundations** – The second Academy course in the Foundations pathway (75–90 min, ChatGPT): repeatable workflows, capability selection and review points. Mirrors the **AP** track. *AP*

**App Server (Codex)** – A Codex surface that hosts Codex capabilities for programmatic and integrated use. Part of the Codex extension/integration surface set. *CX*

**Assessment (Academy)** – The optional-to-take, required-to-pass quiz attached to each Academy course: 10–20 questions drawn from a 50-question bank, ≥ 80% to pass, retakeable with a newly randomised selection. Say "assessment" for the Academy quiz and "mock exam" for ours. *F AP AG API CX L*

**Async tool calling** – Issuing tool calls that resolve asynchronously so the model is not blocked waiting on slow tools. Improves throughput for agentic and multi-tool work. *API AG*

**Auto-review (Codex)** – A Codex capability that reviews changes automatically (for quality and security) before a human sign-off. Supports the review-gate discipline central to Codex governance items. *CX*

## B

**Background mode** – A Responses API mode for running work that continues without a client holding the connection, with results retrieved later. Pairs with streaming and webhooks for long tasks. *API AG*

**Badge** – A digital credential issued via Accredible when an Academy course assessment is passed at ≥ 80%. OpenAI states plainly that badges and pathway certificates of completion are not certifications. *F AP AG API CX L*

**Batch** – An API processing mode for submitting large volumes of requests for asynchronous completion at reduced cost, when latency does not matter. Contrast Flex processing and fast mode. *API*

**Build with Retrieval-Augmented Generation** – The 100-minute Academy API course on retrieval pipelines and answer quality. Part of the Build-with-AI (API) pathway. *API*

## C

**Canvas** – A ChatGPT workspace for iterating on a document or code side-by-side with the chat, rather than in the message stream. Referenced among ChatGPT features for writing and editing. *F AP*

**Chat Completions (legacy)** – The older `/v1/chat/completions` interface. Still supported, but the docs position it as legacy for new work; the Responses API is the primary interface. Choosing Chat Completions for a greenfield build is a common distractor. *API*

**ChatGPT** – OpenAI's consumer and work assistant product and the delivery surface for the Foundations pathway courses and the pilot AI Foundations certification. Models include GPT-5.6 Sol, Sol Pro, Terra, Luna, GPT-5 Thinking Mini, and GPT-6 Pro on higher tiers. *F AP AG*

**ChatGPT Business** – A ChatGPT plan for 2–200 employees with standard and premium seats, no training on business data by default, and admin controls. Premium seats add 5× usage and remove the 5-hour limit. *L F*

**ChatGPT Edu** – The education plan of ChatGPT for institutions, distinct from the free ChatGPT for Teachers offering. *L F*

**ChatGPT Enterprise** – The enterprise plan with custom, credit- or token-based pricing, SAML SSO, SCIM, EKM, RBAC, data residency across ten regions and compliance attestations. The OpenAI Certified app is limited to Enterprise and Edu workspaces, invite-only. *L*

**ChatGPT for Teachers** – Free for verified U.S. K–12 educators through June 2027, with an AFT partnership targeting 400,000 educators. Related to but distinct from the Coursera-based ChatGPT Foundations for Teachers course. *L F*

**ChatKit** – The OpenAI UI toolkit for building chat experiences: customisation, widgets, actions and advanced integrations. Appears in API items about shipping a user-facing surface. *API*

**Chunking** – Splitting documents into passages for indexing in a retrieval pipeline. Chunk size and overlap trade recall against precision and cost; match them to query pattern and document structure. *API*

**Code interpreter** – A sandboxed tool that runs code (for data analysis, file manipulation, computation) inside an agent or ChatGPT session. Prefer it over hallucinated arithmetic for anything numerical. *API AG*

**Codex** – OpenAI's agentic coding system spanning the ChatGPT desktop app, ChatGPT Work on the web, the Codex CLI, the IDE extension, Codex cloud and Codex Micro, plus Slack/GitHub/GitLab/Linear integrations. Mirrors the **CX** track. *CX*

**Codex CLI** – The command-line Codex surface. Key commands: `/model` in-session, `codex --model gpt-5.6` or `codex -m` to select a model, `codex exec -m gpt-5.6 "…"` for non-interactive runs. *CX*

**Codex cloud** – The hosted Codex environment for running tasks with managed compute, internet-access controls and cloud security settings, as opposed to local execution. *CX*

**Codex IDE extension** – The Codex surface embedded in an IDE, sharing the same `config.toml` as the desktop app and CLI. *CX*

**Codex Micro** – A lightweight Codex surface for small, fast tasks within the Codex family. *CX*

**Codex SDK** – A programmatic interface for building on Codex capabilities, part of the extension and integration surface set alongside the App Server and GitHub Action. *CX*

**Codex Security** – The Codex security capability set (plugin, CLI, cloud) providing scans, deep scans, a security workbench, triage, fixes, hardening, vulnerability reports and CI/GitLab CI integration. Central to Codex governance-and-scale items. *CX*

**Company knowledge** – A ChatGPT feature that grounds responses in an organisation's connected internal information. Appears in Applied AI and Leadership items about grounding and governance. *AP L*

**Compaction** – Server-managed summarisation of a long conversation or agent session that preserves the narrative while shrinking token usage. In the Agents API the managed harness performs context summarisation; a similar idea underlies Codex context management. *API AG CX*

**Compliance API** – Programmatic access to compliance logs and audit events for a workspace, used for governed oversight and record-keeping. Distinct from the Admin and Analytics APIs. *L CX API*

**config.toml** – The single shared Codex configuration file for the desktop app, CLI and IDE extension (for example `model = "gpt-5.6"`), covering config basics, advanced settings, environment variables and profiles. *CX*

**Connectors** – Integrations that let a model reach external data and systems, often via MCP. Governance question: what becomes reachable, and under whose permissions. *API AG CX*

**Content provenance** – Signals and practices that record where generated content came from, supporting trust, disclosure and misuse detection. Appears in responsible-use and safety items. *F L API*

**Context management (Codex, experimental)** – An opt-in Codex mode (`features.context_management.experimental_mode = true`) where Astra keeps notes across context windows and can search earlier messages and tool results in the same task. ChatGPT Plus/Pro sign-in only at launch, not Business/Enterprise/API-key. *CX*

**Context window** – The maximum number of tokens a model can attend to at once. The latest API models offer roughly 1.05M tokens; in ChatGPT the window is plan-dependent (for example GPT Instant 54K on Business, 128K on Enterprise; GPT Reasoning 256K). The user-input share is smaller than the total. *F AP API*

**Conversation state** – How a multi-turn exchange is carried forward: managed by the API/harness or by the developer. The Responses API and Agents API offer managed state; owning the loop means owning the state. *API AG*

**Computer use** – A tool family that lets a model operate a graphical interface via screenshots and actions. Sandbox and gate it because it can take real, irreversible actions. *API AG*

**Coursera** – A certification-program partner (learning design) and the platform powering the OpenAI Certified app; the ChatGPT Foundations for Teachers course also launched there. *L*

**Credly** – Credly by Pearson, a certification-program partner (portability) and the distribution channel for OpenAI Certified credentials. Distinct from Accredible, which issues Academy badges. *F AP AG API CX L*

## D

**Data residency** – Where data is stored and processed. ChatGPT Enterprise offers residency across ten regions (US, EU, UK, JP, CA, KR, SG, IN, AU, UAE); the Agents API managed harness is US data residency only. *L API*

**Decomposition** – Breaking a large task into smaller, verifiable steps with clear inputs and outputs. The core Applied AI move that turns one-off prompting into a repeatable workflow. *AP AG*

**Deep research** – A ChatGPT feature that runs longer, multi-step research and produces a synthesised report. Appears in Foundations and Applied AI items about capability selection. *F AP*

**Deep scan (Codex Security)** – A more thorough Codex Security scan that goes beyond a fast scan to find deeper vulnerabilities, feeding triage, fixes and hardening. Part of the Codex Security workbench. *CX*

**Definition of done** – The explicit, checkable conditions that mark a task or workflow step complete. Without it, agents and workflows drift; with it, review gates have something to check against. *AP AG*

**Design and Build Agentic Systems** – The 110-minute Academy API course on agent roles, tools and handoffs. Part of the Build-with-AI (API) pathway. *API AG*

**DPO (Direct Preference Optimization)** – A fine-tuning method that aligns a model to preferred-versus-rejected response pairs. One of the supported fine-tuning types alongside supervised, vision and reinforcement. *API*

## E

**EKM (Enterprise Key Management)** – A ChatGPT Enterprise control that lets an organisation manage its own encryption keys, part of the enterprise security surface alongside SAML SSO, SCIM and RBAC. *L*

**Embeddings** – Numeric vector representations of text used to measure semantic similarity in retrieval. The foundation of a vector store and RAG search. *API*

**Environment (Agents API)** – The OpenAI-hosted or self-hosted sandbox in which an Agent runs: files and artifacts, lifecycle and security. Combined with an Agent and run as a Session. *API AG*

**Environments (Codex)** – Where Codex work runs — modes covering local, cloud and git worktrees — with the associated security and internet-access controls. *CX*

**ETS** – A certification-program partner providing psychometric rigor to the OpenAI Certification initiative. Not involved in the Academy badge assessments. *F AP AG API CX L*

**Evaluate AI Applications** – The 70-minute Academy API course on evals, failure analysis and quality checks. Part of the Build-with-AI (API) pathway. *API*

**Evals** – Systematic evaluation of model or application quality against defined criteria, using datasets and graders. The Evals API is now grouped with legacy APIs, but evals remain conceptually central: getting started, working with evals, external models, best practices. *API*

**Events and items** – The Agents API stream primitives: as a Session runs, it emits events and produces items you follow via streaming or webhooks. Knowing this vocabulary distinguishes managed-runtime items. *API AG*

**Extend Codex Workflows** – The 90-minute Academy Codex course on reusable team workflows and oversight. Part of the Codex pathway. *CX*

## F

**Fast mode** – A latency-optimised API processing mode for when speed matters most. Contrast Batch (cheap, slow) and Flex processing (cheaper, best-effort). *API*

**File search** – A retrieval tool that searches over uploaded files or a vector store to ground responses. One of the built-in tools available to agents. *API AG*

**Fine-tuning** – Adapting a base model to your data or preferences. Supported methods: supervised (SFT), vision, DPO and reinforcement (RFT). Now grouped with legacy APIs but still tested conceptually as a last-resort optimisation after prompting and retrieval. *API*

**Flex processing** – A lower-cost, best-effort API processing mode that trades guaranteed latency for price, sitting between synchronous requests and Batch. *API*

**Foundations pathway** – The Academy pathway combining AI Foundations, Applied AI Foundations and Agents and Workflows. Passing every course assessment at ≥ 80% yields a pathway certificate of completion — which is not a certification. *F AP AG*

**Function calling** – The mechanism by which a model requests a developer-defined function with structured arguments, receives the result, and continues. The foundation of tool use. Pair with structured outputs for reliable arguments. *API AG*

## G

**Get Started with Codex** – The 80-minute Academy Codex course on repository tasks, code changes and review. The entry point of the Codex pathway. *CX*

**GitHub Action (Codex)** – A Codex integration that runs Codex tasks within GitHub Actions CI, part of the non-interactive and automation surface set. *CX*

**GitLab CI (Codex Security)** – Integration that runs Codex Security scans inside GitLab CI pipelines, alongside general CI integration, so vulnerabilities are caught in the pipeline. *CX*

**Gradual** – The platform hosting OpenAI Academy courses. Course, enrolment and progress issues route to Gradual; badge issues route to Accredible. *F AP AG API CX L*

**Graders** – The components in an evals setup that score model outputs against expected results or rubrics. Choosing the right grader (exact match, model-graded, criteria-based) is an evals-design decision. *API*

**GPT-5.3-codex-spark** – A text-only Codex research-preview model (`gpt-5.3-codex-spark`) tuned for near-instant iteration, available on ChatGPT Pro. Distinct from the general GPT-5.6 line-up in Codex. *CX*

**GPT-5.5 / GPT-5.5-pro** – Previous-generation API models still available: `gpt-5.5` at $5 in / $30 out per MTok, `gpt-5.5-pro` at $30 / $180. Terra is the natural replacement for GPT-5.5 workloads. *API CX*

**GPT-5.6 Luna** – `gpt-5.6-luna`, $0.20 in / $1.20 out per MTok, ~1.05M context, 128K max output. The cheapest current model, for clear, repeatable, high-volume tasks: extraction, classification, transformation, structured summaries. *API CX*

**GPT-5.6 Sol** – `gpt-5.6-sol` (alias `gpt-5.6`), $4 in / $20 out per MTok. For complex, open-ended, high-value work. Sol Pro is its ChatGPT counterpart. *API CX*

**GPT-5.6 Terra** – `gpt-5.6-terra`, $2 in / $12 out per MTok. The pragmatic all-rounder and natural replacement for GPT-5.5 workloads. Often the exam-correct default. *API CX*

**GPT-6 Astra** – `gpt-6-astra`, $10 in / $50 out per MTok, reasoning efforts low/medium/high/xhigh/max, ~1.05M context, cutoff 30 Apr 2026. For the hardest end-to-end work: sustained reasoning, judgment, multi-tool. Powers GPT-6 Pro in ChatGPT. *API CX*

**GPT-6 Pro** – The ChatGPT experience powered by GPT-6 Astra, available on Pro $100 / Pro $200 / Business / Enterprise tiers. *F AP AG*

**GPTs** – Custom, shareable ChatGPT assistants configured with instructions, knowledge and tools. Discover, create, share and govern them; workspace controls apply. *F AP L*

**GPT-Live** – OpenAI's real-time voice model family and sessions surface (with delegation and prompting), used alongside the Realtime API for spoken interaction. *API*

**Grounding** – Anchoring a model's output in provided or retrieved source material rather than parametric memory, to reduce hallucination and enable citation. The purpose of RAG, file search and company knowledge. *F AP API*

## H

**Hallucination** – A fluent but false or unsupported model output. Mitigated by grounding, retrieval, verification and human review; never fully eliminated by prompting. The canonical responsible-use concept. *F AP AG API L*

**HIPAA configuration (Codex)** – A managed Codex/ChatGPT Work configuration for handling protected health information under HIPAA, part of the administration surface. Appears in governance items for regulated industries. *CX L*

**Hooks (Codex)** – Configurable trigger points in Codex that run custom logic around agent actions, part of the extending-Codex surface (skills, plugins, MCP, hooks, record and replay). *CX*

**Human-in-the-loop** – Keeping a person in the decision path to review, approve or correct agent actions, especially before irreversible steps. The central oversight pattern of the Agents and Workflows track. *AG AP L*

## I

**IP allowlist** – A network control restricting API or workspace access to approved IP ranges. One of several enterprise access controls alongside Private Link, mTLS and workload identity federation. *API L CX*

**Irreversibility** – The property of an action that cannot be undone (deleting data, sending a message, deploying). High irreversibility raises the bar for a human review gate before the action runs. *AG L CX*

## K

**Knowledge cutoff** – The date after which a model has no built-in knowledge (for example 30 Apr 2026 for Astra, 16 Feb 2026 for the GPT-5.6 line). Anything more recent must be supplied via retrieval or tools. *F API*

## L

**Least privilege** – Granting an agent, tool or user only the minimum permissions needed. Reduces blast radius when something goes wrong; a recurring governance discriminator across the agent and Codex tracks. *AG CX L API*

**Legacy APIs** – The docs grouping that now includes Evals, fine-tuning, Agent Builder and the Assistants API. Refer to these as legacy rather than the newest surfaces, even though evals remain conceptually important. *API*

**Local shell** – A tool that runs shell commands on the local machine within a controlled environment, distinct from a sandboxed remote shell. Requires careful permission scoping. *CX API*

## M

**Max (Codex reasoning)** – A Codex reasoning level meaning more thinking time on a single task. Contrast Ultra, which delegates to parallel subagents. Part of the CLI ladder Low · Medium (default) · High · Extra high · Max · Ultra. *CX*

**MCP (Model Context Protocol)** – The protocol by which models and agents connect to external tools and data sources through MCP servers. Underlies connectors and much of the tool surface across the developer tracks. *API AG CX*

**Memory (ChatGPT)** – A ChatGPT feature that retains user-relevant facts across chats to personalise responses; expanded, with memory across past chats noted as coming. Governance and privacy implications appear in Leadership items. *F L*

**Mid-turn steering** – Adjusting or redirecting a model or agent while it is still working on a turn, rather than only between turns. A Responses/Agents API capability for interactive control of long work. *API AG*

**Misalignment monitoring** – Safety practice of watching for agent or model behaviour that diverges from intended goals, part of the safety-and-governance surface. *API L*

**Mock exam** – This site's independent, full-length practice exam for a track — a diagnostic (mock 1) or readiness gate (mock 2). Never call it "the exam" or an official OpenAI assessment; it is built from publicly available learning objectives. *F AP AG API CX L*

**Moderation** – Classifying content for policy violations (for example unsafe or disallowed material) to filter inputs and outputs. Part of safe deployment. *API L*

**mTLS (mutual TLS)** – Two-way TLS authentication where both client and server present certificates, used as an enterprise transport-security control. *API CX*

**Multi-agent** – Coordinating several agents or subagents on one task. In the Agents API, `multi_agent: { enabled, max_concurrent_subagents }` controls subagent delegation. Codex Ultra applies the same idea in parallel. *API AG CX*

## N

**Non-interactive mode (Codex)** – Running Codex without an interactive session via `codex exec`, suitable for scripts, CI and automation. *CX*

## O

**OpenAI Academy** – OpenAI's free, self-paced learning platform (hosted on Gradual, badges via Accredible) offering 14 courses across three pathways plus standalone courses. Anyone with a ChatGPT account can enrol, globally, free. *F AP AG API CX L*

**OpenAI Certification** – The certification initiative announced 9 December 2025, with AI Foundations delivered inside ChatGPT and a path toward a full certification via additional courses and a hands-on project. A pilot with employer, public-sector and university partners — not a public exam. *F L*

**OpenAI Certified (app)** – A Coursera-powered learning experience inside ChatGPT, with credentials issued via Coursera and distributed through Credly. Limited to ChatGPT Enterprise and Edu workspaces, invite-only. Distinct from Academy badges. *F L*

**OpenAI Jobs Platform** – A jobs-side initiative (with partners including Indeed and Upwork) connecting verified AI skills to work. Context for why the certification initiative exists. *L*

**Optimize AI Application Performance** – The 30-minute Academy API course on quality, speed, reliability and cost in production. Part of the Build-with-AI (API) pathway. *API*

## P

**Pathway** – A grouped sequence of Academy courses (Foundations, Codex, API). Completing every course assessment at ≥ 80% earns a pathway certificate of completion, which OpenAI states is not a certification. *F AP AG API CX L*

**Pathway certificate of completion** – The credential for finishing every course and assessment in an eligible Academy pathway at ≥ 80%. Explicitly not a certification and no guarantee of future eligibility. *F AP AG API CX L*

**Permission modes (Codex)** – Codex settings that determine how much autonomy the agent has to read, edit and run commands before requiring approval. The core lever for balancing speed against safety. *CX*

**Plugins** – Extensions that add capabilities or connect ChatGPT and Codex to external systems; workspace admins control which plugins are available. Part of the ChatGPT and Codex extension surfaces. *F CX AP*

**Predicted outputs** – An API feature that speeds up responses when much of the output is known in advance (for example editing a large file with small changes) by supplying the expected text. A latency-and-cost optimisation. *API*

**Private Link** – A private network connection to OpenAI services that avoids the public internet, used as an enterprise network-security control. *API L*

**Profiles (Codex)** – Named, reusable Codex configuration sets in `config.toml` so different tasks or contexts run with different models, permissions or settings. *CX*

**Programmatic tool calling** – Invoking tools through code-defined logic rather than only model-initiated calls, giving the developer more control over when and how tools run. *API AG*

**Prompt caching** – Reusing the computation for a repeated prompt prefix to cut latency and cost, with cache diagnostics available. Order stable content first so it can be cached. *API*

**Prisma AIRS** – A managed security configuration option referenced in Codex administration for governing AI use in enterprise environments. Appears in Codex governance items. *CX L*

**Projects (ChatGPT)** – A ChatGPT feature for organising related chats, files and instructions into a shared workspace, including shared projects for teams. Referenced among ChatGPT features. *F AP L*

**Prompt optimizer** – An evals-adjacent tool that helps improve a prompt against a task and dataset. Grouped with the evals tooling now under legacy APIs but still practically useful. *API*

**Provenance** – Knowing and being able to show the source of a claim or a piece of content. Enables citation, disclosure and trust; the practical test behind grounding and content-provenance controls. *F AP L API*

## R

**Rate limits** – Ceilings on request or token throughput per unit time, enforced per organisation or project. Handle with backoff and by choosing the right processing mode. *API*

**RBAC (role-based access control)** – Granting permissions by role rather than to individuals, in ChatGPT Enterprise, Codex workspaces and the API platform. The scalable expression of least privilege. *L CX API*

**Readiness band** – This site's independent readiness indicator, never a score: under 70% keep learning, 70–79% building confidence, 80–89% assessment ready, 90%+ strong readiness. The 80% line matches the Academy badge threshold. *F AP AG API CX L*

**Realtime API** – The API for low-latency spoken and streaming interaction, supporting voice activity detection, tools and MCP, and transport over WebRTC, WebSockets or SIP. Backed by the realtime model family. *API*

**Reasoning effort** – A control on how much internal reasoning a model spends. API efforts run none through max (Astra adds xhigh); the guidance is to use the lowest effort that gets the result. Codex exposes an analogous CLI ladder. *API CX*

**Record and replay (Codex)** – A Codex capability to capture and re-run agent interactions, useful for debugging and reproducibility, part of the extending-Codex surface. *CX*

**Record mode (ChatGPT)** – A ChatGPT feature that captures a session (for example a meeting) as input for later processing. Referenced among ChatGPT features in the non-developer tracks. *AP F*

**Red teaming** – Adversarial testing of a model or application to surface unsafe or unintended behaviour before release. Part of the safety-and-governance surface and responsible-deployment items. *API L*

**Reranking** – A second-stage retrieval step that reorders first-stage candidates by relevance to improve precision before the model reads them. A standard lever for RAG answer quality. *API*

**Responses API** – The primary OpenAI API interface (the recommended surface for new work over legacy Chat Completions), served across the model line-up, supporting conversation state, background mode, streaming, tools, compaction and more. *API AG*

**Retrieval** – Fetching relevant external information to ground a response, as opposed to relying on model memory. The subject of the Build with Retrieval-Augmented Generation course and RAG items. *API AP*

**Retrieval-Augmented Generation (RAG)** – A pattern that retrieves relevant passages and supplies them to the model as context to ground its answer, improving accuracy and enabling citation. The pipeline: chunk, embed, store, retrieve, optionally rerank, generate. *API AP*

**Review gate** – An explicit checkpoint where output must be verified or approved before proceeding, especially before irreversible actions. The operational form of human-in-the-loop in Applied AI and agent workflows. *AP AG CX L*

**RFT (Reinforcement Fine-Tuning)** – Fine-tuning that optimises a model against a reward or grader signal on your task. One of the supported fine-tuning methods; more involved than SFT or DPO. *API*

**Rules (Codex)** – Configurable constraints that govern how the Codex agent behaves, defined in agent configuration alongside AGENTS.md and subagents. *CX*

## S

**Safety classifiers** – Models that classify content or behaviour for safety risks, used to filter and monitor inputs, outputs and agent actions. Part of the safety-and-governance surface. *API L*

**SAML SSO** – Single sign-on via SAML for ChatGPT Enterprise and workspaces, with SCIM for user provisioning. Core enterprise identity controls. *L*

**Scale Codex Across Governed Teams and Systems** – The 100-minute Academy Codex course on parallel workstreams, coordination and governance. The capstone of the Codex pathway. *CX*

**SCIM** – A protocol for automated user provisioning and de-provisioning in ChatGPT Enterprise, paired with SAML SSO. Supports user-lifecycle governance. *L*

**Sandboxing** – Running code, commands or agents in an isolated environment so mistakes or attacks cannot affect the host. Codex sandboxing (including Windows sandbox and WSL) and hosted Agents API environments are the concrete forms. *CX API AG*

**Scope AI Solutions** – The 30-minute Academy API course on identifying AI opportunities and planning solutions. Part of the Build-with-AI (API) pathway. *API L*

**Secure MCP tunnel** – A secured channel for connecting an agent to an MCP server, part of the tool-and-connector surface, letting tools reach systems that are not publicly exposed. *API AG*

**Sessions (Agents API)** – A durable running instance of an Agent in an Environment: create it, give it a task, follow progress via streaming or webhooks, then continue or steer it. The unit that OpenAI manages for you. *API AG*

**SFT (supervised fine-tuning)** – Fine-tuning on labelled input/output examples to teach a model a specific format or task. The most common fine-tuning method; contrast DPO and RFT. *API*

**Shell (tool)** – A tool that lets an agent run shell commands, either locally or in a sandbox, to accomplish real work. Powerful and dangerous — scope permissions and gate destructive commands. *CX API AG*

**Site tools (WebMCP)** – A Codex extension mechanism that exposes website tools to the agent via MCP, part of the extending-Codex surface. *CX*

**Sites (ChatGPT)** – A ChatGPT feature for building and publishing simple sites from within the product. Referenced among ChatGPT features in the non-developer tracks. *F AP*

**SIWC (Sign in with ChatGPT)** – Signing into Academy and other surfaces with a ChatGPT account; progress and badges bind to the email used, and accounts cannot be merged after a course starts. *F AP AG API CX L*

**Skills** – Reusable, packaged capabilities that extend what a model or agent can do; available in ChatGPT (beta), the API tool surface and Codex. Referenced across tracks as a way to standardise repeated work. *API CX AP*

**Spend limits** – Configurable ceilings on spending per organisation or project, a governance control to prevent runaway cost. Pairs with rate limits and monitoring. *API L*

**Streaming** – Returning a response incrementally as it is generated rather than all at once, improving perceived latency and enabling live UI and mid-turn steering. *API AG*

**Structured outputs** – Constraining a model's response to a defined schema (for example JSON) so it can be parsed reliably. The exam-correct answer whenever downstream code must consume the output. Pair with function calling. *API AG*

**Study mode** – A ChatGPT feature that guides a learner through material with questions and explanations rather than just answers. Referenced in Foundations and education contexts. *F*

**Subagents** – Delegated agents that a parent agent spawns to handle parts of a task, in the Agents API (`multi_agent`) and Codex (Ultra). Enable parallelism at the cost of coordination overhead. *API AG CX*

**Sycophancy** – A model's tendency to agree with or flatter the user rather than give an accurate or critical answer. A responsible-use failure mode to watch for, distinct from hallucination. *F AP L*

## T

**Terraform provider** – An infrastructure-as-code provider for managing OpenAI platform resources declaratively, part of the infra-and-ops surface. Appears in API production and governance items. *API*

**Token** – The unit of text a model processes and is billed on (roughly a word-piece). Pricing is per million tokens (MTok), and context windows are measured in tokens. The foundational cost-and-limit concept. *F AP API*

**Tool search** – A capability that lets an agent find the right tool from a large set rather than being handed a fixed list, useful when many tools are available. Part of the tool surface. *API AG*

## U

**Ultra (Codex reasoning)** – The top Codex reasoning level: automatic delegation to subagents working in parallel. Enable it via Settings → Configuration → "Ultra in model picker slider" if absent. Contrast Max, which is more thinking on one task. *CX*

## V

**Vector store** – An index of embeddings that supports fast semantic similarity search over documents, backing file search and RAG. Choosing chunking and what to store are its key design decisions. *API*

**Vision (fine-tuning)** – Fine-tuning that adapts a model on image inputs, one of the supported fine-tuning methods alongside SFT, DPO and RFT. All latest API models already accept image input by default. *API*

**Vision (model capability)** – The ability of the latest models to accept image input alongside text. Enables document understanding, screenshots and computer use. *API AP*

## W

**Web search** – A built-in tool that lets a model retrieve current information from the web, filling the gap left by a knowledge cutoff. One of the core tools for grounding. *API AG AP*

**WebRTC / WebSockets / SIP** – Transport options for the Realtime and voice APIs: WebRTC and WebSockets for low-latency streaming, SIP for telephony integration. Referenced in voice/realtime items. *API*

**Webhooks** – HTTP callbacks that notify your system when an event occurs (for example an Agents API session emitting an item or completing), so you do not have to poll. Pair with background mode. *API AG*

**Workflow** – A predefined sequence of steps with explicit inputs, outputs and checks, as opposed to an agent that decides its own path. Applied AI is largely about designing reliable workflows before reaching for agents. *AP AG*

**Workflow contract** – The agreed inputs, outputs, constraints and acceptance criteria for a workflow step, so each stage can be built and verified independently. The Applied AI discipline that makes decomposition reliable. *AP AG*

**Workload identity federation** – Authenticating workloads to OpenAI without long-lived secrets, using federated identity across K8s, AWS, Azure, GCP, OCI, GitHub Actions, SPIFFE and X.509. A production security control. *API CX*

**Worktrees (git)** – Multiple working directories for one repository, letting Codex run parallel tasks on different branches without collisions. Part of Codex environment modes. *CX*

**Workspace agents** – A ChatGPT feature for agentic work inside a workspace, referenced among the ChatGPT surfaces the non-developer tracks cover. *AP AG*

## Z

**ZDR (Zero Data Retention)** – A configuration in which request data is not retained. Note the constraint: the Agents API managed harness does not offer ZDR (and a self-hosted sandbox does not change that), which can rule it out for sensitive workloads. *API L*