OpenAI

What OpenAI does

OpenAI states its mission as ensuring that artificial general intelligence (AGI) benefits all of humanity.6367

• OpenAI builds AI models and offers them in two main ways: ChatGPT and Codex plans for people and organizations, and API access for developers, billed by usage at API rates.75
• ChatGPT is its assistant for questions, research, writing, files, images, and voice. The same app can also take on longer tasks through ChatGPT Work and coding tasks through Codex.30
• The OpenAI API gives developers hosted models for text, images, speech, transcription, and embeddings, plus tools for building AI agents such as the Responses API, the Agents SDK, and the Agents API.171525
• OpenAI publishes some code and model weights openly. The Codex CLI is Apache-2.0 code on GitHub, gpt-oss-120b and gpt-oss-20b are Apache 2.0 open-weight models, and its Hugging Face page also hosts the gpt-oss-safeguard models and an Apache 2.0-licensed Privacy Filter model.891080
• For organizations it sells Business, Enterprise, and Edu plans, and it has made public-sector agreements such as an August 2025 GSA OneGov deal that offered ChatGPT Enterprise to participating federal agencies.772
• OpenAI models are also offered through other clouds. OpenAI's changelog says they became available in Amazon Bedrock through an OpenAI-compatible Responses API endpoint on June 1, 2026.15
• Some models are offered only to approved organizations: GPT-Rosalind for life sciences research and GPT-5.6 Cyber for authorized security testing.165222

ChatGPT in depth

OpenAI describes ChatGPT as "an AI agent that you communicate with in natural language." People start with a question, idea, file, or task, add context and tools, then review and refine the result.30

• ChatGPT has three modes. Chat is for questions, brainstorming, drafting, and summarizing. ChatGPT Work is for bigger tasks that end in a result to review, such as a slide deck, report, or project plan. Codex is for software work such as debugging, running tests, and reviewing pull requests.30
• ChatGPT can search the web and compare sources. Its deep research feature, described in a February 2025 system card, runs multi-step research across web pages, images, and PDFs and can analyze data by writing and running Python code. It ran on an early version of OpenAI o3 and was available to Pro subscribers at launch.3061
• ChatGPT agent, introduced in July 2025, combined deep research with Operator, an earlier tool that worked in a remote web browser. It added a terminal for running code and making slides or spreadsheets, and could connect to services such as Google Drive.60
• ChatGPT Work, launched in July 2026, gathers context from files and plugins, works toward a goal in steps, and produces documents, presentations, spreadsheets, and Sites for review. It can run in a managed cloud environment or, from the desktop app, use local files, apps, and the browser. Scheduled tasks can repeat or refresh work.2932
• Dots, rolling out gradually since late September 2026, are always-on agents powered by GPT-6 Astra. Each runs in the cloud with its own computer and browser, keeps track of ongoing responsibilities, and can be reached in ChatGPT, Slack, or Microsoft Teams. The docs list them for Pro, Business Premium, and Enterprise plans; on Pro they are limited to people over 18 outside the EEA, UK, and Switzerland.3729
• Memories carry useful context from earlier chats into later ones and are managed in Settings, under Personalization. On macOS, an opt-in feature called Computer History turns activity in allowed apps and websites into a searchable timeline and memories; OpenAI says it does not include screenshots or audio, and it can be paused or cleared.3441
• Projects group related chats, files, and instructions. ChatGPT Space, added in late September 2026, holds Pages that people write and edit together with ChatGPT, along with spreadsheets, presentations, Sites, and files that can be shared with collaborators.353829
• ChatGPT Voice is powered by GPT-Live, a voice model that can listen and respond at the same time. The GPT-Live system card says GPT-Live-1 is the default voice model on paid plans and GPT-Live-1 mini on the free plan. In the desktop app, Voice can start, check on, and steer other tasks.3648
• OpenAI added image generation built into GPT-4o in March 2025, describing it as a different approach from its earlier DALL·E models. ChatGPT Images 2.0 followed in April 2026 and ChatGPT Images 2.5 in September 2026; OpenAI says Images 2.5 adds C2PA provenance metadata and an invisible SynthID watermark.515049
• Plugins connect ChatGPT and Codex to outside tools such as Google Drive, SharePoint, Slack, and GitHub. Developers build them from MCP servers, reusable skills, and optional on-screen interfaces, and can submit them to a ChatGPT directory.3026
• Sites, in public beta since June 2026, lets ChatGPT build, host, and share websites, web apps, and games. A new Site is visible only to its owner and workspace admins until its access is changed. Sites is included with Plus, Pro, Business, Enterprise, and Edu plans.3929
• A ChatGPT browser extension for Chrome, Edge, Brave, Opera, and Vivaldi lets the desktop app work inside a person's browser. ChatGPT asks before using each new site, and browser history access is a separate permission it requests each time.40

Plans, apps, and data controls

OpenAI's pricing documentation lists ChatGPT plans named Free, Go, Plus, Pro, Business, and Enterprise and Edu, plus an option to use Codex with an API key.7

• Free and Go include lighter Codex use with GPT-6 Luna in the desktop app. Plus adds Codex on the web, in the CLI, in the IDE extension, and on iOS, along with GPT-6.1 Sol. Pro includes everything in Plus with several usage tiers, and Ultrafast mode is offered on the highest Pro tier.7
• Business adds a dedicated workspace with admin controls, SAML single sign-on, and multi-factor authentication, and the pricing page says Business data is not used for training by default. Enterprise and Edu add SCIM, enterprise key management, role-based access control, audit logs through the Compliance API, and data retention and residency controls.7
• Model choice depends on the plan. In August 2026 GPT-5.6 Luna became the default on Free and Go, while Plus and Pro got a slider for how long GPT-5.6 Sol thinks. On October 7, 2026, OpenAI said GPT-6 Sol and GPT-6 Luna were launching to everyone in ChatGPT and would replace GPT-5.6 Sol and Luna there.2945
• For ChatGPT Work and Codex, the model docs list GPT-6 Astra, GPT-6.1 Sol, GPT-6 Sol, and GPT-6 Luna. GPT-6.1 Sol is rolling out on Plus, Pro, Business, Enterprise, and Edu, and stays off for Enterprise and Edu until an admin enables it. GPT-5.5 is scheduled to leave ChatGPT, ChatGPT Work, and Codex on October 14, 2026, while remaining in the API.337
• ChatGPT works on the web at chatgpt.com without installing anything, in a desktop app for macOS, Windows, and Linux, and in a mobile app. The Linux app entered preview on August 11, 2026 for supported Ubuntu, Debian, and Fedora systems.3129
• On July 9, 2026, the Codex app merged into the ChatGPT desktop app for macOS and Windows. OpenAI says the desktop app is available on every plan, including Free.29
• ChatGPT can be brought into Slack and Microsoft Teams conversations, and app developers can offer "Sign in with ChatGPT" so people can use their ChatGPT plan in other apps; commercial use of that sign-in was in a limited trial as of DevDay 2026.29
• In ChatGPT's settings, people can view and revoke shared links under data controls, manage or turn off memories under Personalization, and pause, filter, or clear Computer History.303441
• For the API, OpenAI says data sent to the API has not been used to train its models by default since March 1, 2023. Abuse-monitoring logs are kept for up to 30 days, and approved customers can get Zero Data Retention.23
• OpenAI's developer docs also cover advertising and shopping in ChatGPT: an Ads API for advertisers to create and measure ads in ChatGPT, and the Agentic Commerce Protocol, which merchants and payment providers use to support checkout inside ChatGPT.2728

Models over time

OpenAI's API changelog, model list, and deprecations page record when each model arrived and when older ones are switched off.151718

• OpenAI published its GPT-4 Technical Report on March 15, 2023. GPT-3.5 Turbo, GPT-4, and GPT-4 Turbo are now deprecated, with an API shutdown date of October 23, 2026 and GPT-5.6 models named as replacements.7718
• GPT-4o arrived in the API on May 13, 2024 and GPT-4o mini on July 18, 2024. Both are still listed, but the dated gpt-4o-2024-05-13 snapshot shuts down on October 23, 2026, and the chatgpt-4o-latest alias was shut down on February 17, 2026.151718
• The o-series reasoning models, which work through a problem before answering, began with o1-preview and o1-mini on September 12, 2024, followed by o1 (December 17, 2024), o3-mini (January 31, 2025), and o3 and o4-mini (April 16, 2025). o1, o3-mini, and o4-mini are scheduled to leave the API on October 23, 2026.1518
• GPT-4.5 Preview (February 27, 2025) was shut down on July 14, 2025, with GPT-4.1 as its replacement. The GPT-4.1 family arrived on April 14, 2025; GPT-4.1 and GPT-4.1 mini are still listed, while GPT-4.1 nano shuts down on October 23, 2026.151817
• GPT-5 arrived on August 7, 2025. Its system card describes it in ChatGPT as one system with a fast model, a deeper reasoning model, and a router that picks between them, taking over from GPT-4o, GPT-4o mini, o3, o4-mini, and o3 Pro.5815
• Updates followed quickly: GPT-5.1 Instant and Thinking (November 2025), GPT-5.2 (December 11, 2025), GPT-5.3 Instant (March 2, 2026), GPT-5.4 and GPT-5.4 Pro with a 1M-token context window (March 5, 2026), GPT-5.5 and GPT-5.5 Pro (April 24, 2026), and GPT-5.5 Instant (May 5, 2026).57155654
• GPT-5.6, released on July 9, 2026 after a limited preview in late June, introduced three sizes named Sol, Terra, and Luna. The system card describes Sol as the main model, Terra as a capable lower-cost option, and Luna as a quicker, more economical option.464715
• The GPT-6 series followed: GPT-6 Astra on September 3, 2026, GPT-6 Sol and GPT-6 Luna on September 22, and GPT-6.1 Sol on September 29. The API pages list a 1,050,000-token context window and up to 128,000 output tokens for these models, with text and image input and text output.15192021
• Two specialized lines require approval: GPT-5.6 Cyber, added on August 7, 2026 for authorized vulnerability research through the Daybreak program, and GPT-Rosalind for life sciences research, generally available through trusted access from September 8, 2026. GPT-Rosalind-5.5 was trained further from GPT-5.5.221552
• Open-weight releases include the Whisper speech recognition models (from September 2022), gpt-oss-120b and gpt-oss-20b (August 5, 2025), gpt-oss-safeguard (October 29, 2025), and Privacy Filter, a model for finding and masking personal information with 1.5B total parameters, about 50M of them active per token.7891580
• Image models moved from DALL·E 2 and 3, removed from the API on May 12, 2026, to gpt-image-1 (April 23, 2025), gpt-image-2 (April 21, 2026), and GPT Image 2.5 Sunburst and Flare (September 8, 2026). The Sora 2 video model launched in September 2025; OpenAI's deprecations page lists the Videos API and Sora 2 models as shut down on September 24, 2026.15186223
• Audio models include the realtime voice models (including GPT-Realtime-2.1, released July 6, 2026), GPT-Live 1, and GPT-Transcribe and GPT-Live-Transcribe (July 28, 2026). whisper-1 and the GPT-4o transcription models are set to shut down on February 26, 2027. The text-embedding-3 models (January 25, 2024) and omni-moderation (September 26, 2024) are still listed.151817

Codex

OpenAI's Codex README calls Codex CLI "a coding agent from OpenAI that runs locally on your computer," and points to Codex in code editors, a desktop app, and Codex Web, a cloud-based agent at chatgpt.com/codex.8

• OpenAI launched the Codex CLI on April 16, 2025, alongside o3 and o4-mini. It is open source under the Apache-2.0 license, installs by script, npm, or Homebrew, and signs in with a ChatGPT plan (Plus, Pro, Business, Edu, or Enterprise) or an API key.158
• Codex used its own Codex-specific models for a time: codex-mini-latest (May 15, 2025), GPT-5-Codex (September 23, 2025), GPT-5.1-Codex (November 13, 2025), GPT-5.1-Codex-Max (December 4, 2025), GPT-5.2-Codex (January 14, 2026), and GPT-5.3-Codex (February 24, 2026). Most are now deprecated, and GPT-5.3-Codex is set to leave the API on April 1, 2027.1518
• Codex now runs OpenAI's general models. Codex CLI 0.161.0, published on October 7, 2026, made GPT-6.1 Sol the default model in its bundled catalog.8333
• The Codex desktop app arrived on macOS in early February 2026 and on Windows in early March 2026. On July 9, 2026 it merged into the ChatGPT desktop app, adding inline diff editing, pull request review, and projects that span several repositories.29
• Through 2026 Codex added plugins (late March), generally available scheduled tasks (March 12), browser control and macOS computer use (April), Windows computer use (late May), and Remote, which lets people continue desktop work from the mobile app (May).29
• Codex cloud features include reusable cloud environments and integrations with GitHub, Slack, Linear, and GitLab (in beta since August 19, 2026), including automatic code review on paid plans.6297
• Codex Security reviews pull requests for security problems (research preview from August 6, 2026), and Codex Security Cloud, announced at DevDay 2026, scans connected repositories and prepares patches, also as a research preview.29
• Codex Micro, launched on July 15, 2026 with Work Louder, is a limited-run physical control surface for Codex with status keys, an analog stick, and a dial for adjusting reasoning effort.29
• Codex can use OpenAI models through Amazon Bedrock with AWS-managed sign-in and billing (June 2026), and developers can build on the Codex harness itself through the Agents API (public beta from September 10, 2026).292515
• GPT-5.3-Codex-Spark, a near-instant research-preview model for real-time coding added in February 2026, was retired from Codex on September 14, 2026.2933

OpenAI API and developer platform

The OpenAI API serves hosted models through endpoints such as Responses, Chat Completions, Batch, and Realtime. On October 6, 2026 OpenAI cut its usage tiers from five to three, named Build, Launch, and Grow.51915

• The Responses API, released on March 11, 2025, is OpenAI's API "for creating and using agents and tools." Built-in tools added since include web search, file search, computer use, remote MCP servers, a code interpreter, a hosted shell, skills, and tool search.1519
• The Conversations API (August 20, 2025) stores long-running conversations. The older Assistants API, introduced on November 6, 2023, shut down on August 26, 2026, with the Responses and Conversations APIs as replacements.1518
• The Agents SDK, released with the Responses API in March 2025 and in TypeScript in June 2025, is OpenAI's open-source library for building agents. 2026 updates added sandboxes and a built-in open-source harness.15
• The Agents API entered public beta on September 10, 2026. It gives applications the Codex harness as an OpenAI-managed service with lasting sessions and hosted or self-hosted sandboxes; computer use was added on September 29. The docs say it supports data residency only in the United States and does not offer Zero Data Retention.2515
• The Realtime API for speech-to-speech apps was introduced on October 1, 2024 and became generally available on August 28, 2025. GPT-Live 1, a full-duplex voice model, became generally available in the API on September 10, 2026.15
• The Decisions API, released in beta on October 6, 2026 with gpt-6-luna, turns text and images into typed answers.15
• Agent Builder and ChatKit arrived at DevDay on October 6, 2025. On June 3, 2026 OpenAI announced that Agent Builder, the Evals platform, and reusable prompts will shut down on November 30, 2026; ChatKit stays, and the Agents SDK or ChatGPT workspace agents are the suggested paths.1518
• On January 15, 2026 OpenAI announced Open Responses, an open-source specification for building interfaces that work with many model providers, based on the Responses API.15
• Developers extend ChatGPT and Codex with plugins built on MCP servers. At DevDay 2026 OpenAI added Plugin Extensions for sidebar apps, panels, editors, and forms, and MCP Events, which let ChatGPT act on updates from an MCP server.2629
• Enterprise controls added over 2025 and 2026 include customer-managed encryption keys (October 24, 2025), hard spend limits (July 22, 2026), mutual TLS and workload identity federation (August 29, 2026), and an in-product flow for HIPAA business associate agreements (October 5, 2026).15

Research, safety, and policies

OpenAI's Deployment Safety Hub collects system cards for its launches, including GPT-5, GPT-5.6, GPT-6 Astra, GPT-Live, ChatGPT agent, deep research, Sora 2, and ChatGPT Images.42

• Version 2 of OpenAI's Preparedness Framework, dated April 15, 2025, tracks three areas: biological and chemical, cybersecurity, and AI self-improvement. OpenAI says it will not deploy a model that reaches a High capability threshold until the related risks are sufficiently reduced, and that reaching a Critical threshold also requires safeguards during development.63
• Under the framework, an internal Safety Advisory Group recommends safeguards, OpenAI leadership approves or rejects them, and the board's Safety and Security Committee oversees those decisions. Version 2 also lists research areas such as long-range autonomy, sandbagging, autonomous replication, undermining safeguards, and nuclear and radiological risk. The GPT-6 Astra system card links to this version.6343
• OpenAI treated ChatGPT agent (July 2025) and GPT-5's thinking model as High capability in biology and chemistry. It treated GPT-5.5 as High in cybersecurity but below Critical, and the GPT-5.6 models as High in both areas; none reached High in AI self-improvement.60585346
• The GPT-6 Astra system card (September 3, 2026) says Astra reaches the Critical level in cybersecurity and is treated as High in biology and chemistry. A September 29 addendum treats GPT-6.1 Sol as Critical in cybersecurity too. Listed safeguards include misalignment monitoring on all tool-using requests and real-time misuse monitors.4344
• Sensitive capabilities go through approval-based programs: Trusted Access for Cyber, the Daybreak Blue and Daybreak Red tiers for security defenders (August 2026), and Trusted Access for Biology Research. GPT-Rosalind is offered only to approved customers after a review of their mission, governance, and access controls.431552
• The Model Spec describes how OpenAI intends its models to behave. Its original version is dated May 8, 2024; it has been public-domain (CC0) source on GitHub since February 12, 2025. Later versions added agent principles (September 12, 2025), an Under-18 Safety Mode (December 18, 2025), and clarifications on August 18, 2026.646566
• For teens, the Model Spec marks rules for people under 18 with a "U18" badge. The GPT-6 Astra card describes expanded under-18 protections, and the Sora 2 card (2025) said parental controls were being introduced. OpenAI's API guidance for apps serving minors requires following child-protection laws such as COPPA.66436224
• An October 27, 2025 addendum covers how ChatGPT handles conversations about mental health, emotional reliance, and self-harm. It describes an October 3 update to the default model meant to respond with care and point people toward real-world support.59
• OpenAI marks generated media with C2PA provenance metadata. Sora 2 videos carried a visible moving watermark, and ChatGPT Images 2.5 adds an invisible SynthID watermark in ChatGPT, Codex, and the API.516249
• API safety tools include moderation scores in the Responses and Chat Completions APIs (June 4, 2026) and a Safety Usage Dashboard (June 23, 2026). The open-weight gpt-oss-safeguard models, released on October 29, 2025, classify text against safety policies that the developer provides.1579
• OpenAI's usage policies and terms apply to developers. Its under-18 guidance adds that apps must not process data from children under 13 without Zero Data Retention, and that OpenAI may audit organizations and suspend access for repeated or serious noncompliance.24

Infrastructure, compute, and government work

A joint release from OpenAI, Oracle, and SoftBank describes Stargate as OpenAI's AI infrastructure platform.74

• The Stargate Project was announced in Washington, D.C. on January 21, 2025 as a new company to build AI infrastructure for OpenAI in the United States, starting in Texas. SoftBank holds financial responsibility and OpenAI operational responsibility; SoftBank, OpenAI, Oracle, and MGX are the initial equity funders, and Masayoshi Son chairs it.73
• Initial technology partners named for Stargate were Arm, Microsoft, NVIDIA, Oracle, and OpenAI.73
• On September 23, 2025 the partners named five more U.S. sites: Shackelford County, Texas; Doña Ana County, New Mexico; and a Midwest site, developed by Oracle; and Lordstown, Ohio, and Milam County, Texas, developed with SoftBank, with SB Energy supplying power infrastructure for Milam County.74
• The same release said the main campus in Abilene, Texas, was running on Oracle Cloud Infrastructure with NVIDIA GB200 racks delivered from June 2025, and that the announced sites, Abilene, and CoreWeave projects brought Stargate to nearly 7 gigawatts of planned capacity.74
• For API customers, OpenAI offers data storage and processing in the United States, Europe, and the United Arab Emirates, and storage only in Australia, Canada, Japan, India, Singapore, South Korea, and the United Kingdom; regions outside the United States require extra approvals.23
• On August 6, 2025 the U.S. General Services Administration announced a OneGov agreement making ChatGPT Enterprise available to participating federal agencies for one year at a nominal cost, plus training resources; GSA said it supports OMB memoranda M-25-21 and M-25-22.72
• OpenAI said it previewed GPT-5.6 with the U.S. government and, at the government's request, started with a limited preview for a small group of trusted partners on June 26, 2026 before a wider release.47
• GPT-Rosalind has a separate trusted-access path for government applicants working on biosecurity, public health, and biodefense.52
• On October 8, 2026 a White House fact sheet listed OpenAI among eleven industry partners committing AI-for-science tools and compute credits to the Genesis Mission Consortium.75

Corporate structure and governance

A February 2026 agreement filed with the SEC describes OpenAI Group PBC as "a Delaware public benefit corporation" at 1455 3rd Street, San Francisco; it was signed for the company by Sarah Friar, Chief Financial Officer.1

• OpenAI, Inc. was incorporated in 2015 as a Delaware nonprofit.2
• On October 28, 2025 the Delaware attorney general issued a Statement of No Objection to OpenAI's recapitalization. OpenAI had initially proposed ending the nonprofit's oversight of its for-profit work, but instead agreed to create a public benefit corporation that the nonprofit controls; the nonprofit alone can appoint and remove the PBC's directors.2
• The PBC's mission is identical to the nonprofit's, and on safety and security matters its directors must consider only the mission. The nonprofit also gets access to the PBC's models and APIs.2
• The Safety and Security Committee stays with the nonprofit and can require safety measures "up to and including halting the release of models." Its chair sits on the nonprofit board but not the PBC board, with rights to observe PBC board meetings.2
• Under an October 27, 2025 memorandum of understanding with the California attorney general, the nonprofit and the PBC will keep their headquarters in California.3
• The nonprofit now operates as the OpenAI Foundation. It describes itself as a separate nonprofit corporation that runs independently for its charitable mission, and says "The Foundation controls OpenAI Group PBC."7169
• The Foundation lists three initial program areas: life sciences and curing diseases, AI resilience, and civil society and philanthropy. It also runs a community grant fund, whose 2026 grantees it announced on October 1, 2026.6768
• Bret Taylor chairs both the Foundation and OpenAI Group PBC boards. David Vélez and Robin Vince joined both boards on July 21, 2026, and Paul Christiano joined the Foundation board and its Safety and Security Committee on September 9, 2026.7069

Recent updates (July to October 2026)

• July 6-10, 2026: OpenAI launched ChatGPT Work. On July 9 it released the GPT-5.6 family (Sol, Terra, Luna) and merged the Codex app into the ChatGPT desktop app.2915
• July 8, 2026: OpenAI published the GPT-Live system card. In late July it introduced ChatGPT Voice powered by GPT-Live in the desktop app, which can start and steer tasks.4829
• July 15 and 21, 2026: OpenAI released Codex Micro, a limited-run control surface, and David Vélez and Robin Vince joined the boards of the OpenAI Foundation and OpenAI Group PBC.2970
• August 7-11, 2026: OpenAI added GPT-5.6 Cyber and the Daybreak Blue and Red access tiers for approved security defenders, and opened a preview of the ChatGPT desktop app for Linux.1529
• August 26, 2026: The Assistants API shut down, and OpenAI announced that whisper-1 and the GPT-4o transcription models will shut down on February 26, 2027.1518
• September 3, 2026: OpenAI released GPT-6 Astra and its system card, which says the model reaches the Critical level in cybersecurity under the Preparedness Framework.1543
• September 8-10, 2026: OpenAI released GPT Image 2.5 Sunburst and Flare, made GPT-Rosalind generally available through trusted access, added Paul Christiano to the Foundation board (September 9), and opened the Agents API in public beta (September 10).154969
• September 22-24, 2026: OpenAI released GPT-6 Sol and GPT-6 Luna, and September 24 was the listed shutdown date for the Videos API and Sora 2 models.1518
• September 29, 2026: At DevDay in San Francisco, OpenAI released GPT-6.1 Sol and announced dots, ChatGPT Space, ChatGPT in Slack and Teams, Team Tasks, Plugin Extensions, and computer use in the Agents API.811529
• October 6-8, 2026: OpenAI released the Decisions API in beta (October 6), began launching GPT-6 to everyone in ChatGPT (October 7), and made GPT-6.1 Sol the default in Codex CLI 0.161.0 (October 7); on October 8 the White House listed OpenAI as a Genesis Mission Consortium partner.15458375
• Scheduled: GPT-5.5 leaves ChatGPT, ChatGPT Work, and Codex on October 14, 2026, and GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, o1, o3-mini, o4-mini, and gpt-image-1 are set to leave the API on October 23, 2026.3318

Models and products lineup

Newest first. Status is as the official source describes it; “Legacy” and “Retired” entries stay listed for history.

Timeline

1. 2015

   OpenAI, Inc. is incorporated as a Delaware nonprofit.2

2. Sep 2022

   OpenAI releases the original Whisper speech recognition models.78

3. Nov 2022

   OpenAI launches ChatGPT.76

4. Mar 1, 2023

   OpenAI stops using data sent to its API for model training by default.23

5. Mar 15, 2023

   OpenAI publishes the GPT-4 Technical Report.77

6. Nov 6, 2023

   At DevDay, OpenAI introduces GPT-4 Turbo in preview, the Assistants API, and DALL·E 3 and text-to-speech in the API.15

7. May 8, 2024

   OpenAI publishes the original version of its Model Spec.65

8. May 13, 2024

   GPT-4o becomes available in the API.15

9. Sep 12, 2024

   OpenAI releases its o1-preview and o1-mini reasoning models.15

10. Jan 21, 2025

    The Stargate Project, a new company to build AI infrastructure for OpenAI, is announced in Washington, D.C.73

11. Mar 11, 2025

    OpenAI releases the Responses API and the Agents SDK.15

12. Mar 25, 2025

    OpenAI publishes the system card for image generation built into GPT-4o.51

13. Apr 15, 2025

    OpenAI updates its Preparedness Framework to version 2.63

14. Apr 16, 2025

    OpenAI releases o3 and o4-mini and launches the Codex CLI.15

15. Jul 17, 2025

    OpenAI publishes the ChatGPT agent system card for its new agent in ChatGPT.60

16. Aug 5, 2025

    OpenAI releases the open-weight gpt-oss-120b and gpt-oss-20b models.9

17. Aug 6, 2025

    GSA announces a OneGov agreement offering ChatGPT Enterprise to participating federal agencies.72

18. Aug 7, 2025

    OpenAI releases GPT-5 in ChatGPT and the API.5815

19. Sep 23, 2025

    OpenAI, Oracle, and SoftBank name five more U.S. Stargate data center sites.74

20. Sep 30, 2025

    OpenAI publishes the system card for Sora 2, its video and audio generation model.62

21. Oct 28, 2025

    The Delaware attorney general issues a Statement of No Objection to OpenAI's recapitalization into a public benefit corporation controlled by the nonprofit.2

22. Nov 13, 2025

    GPT-5.1 becomes available in the API.15

23. Dec 11, 2025

    OpenAI releases GPT-5.2.15

24. Dec 18, 2025

    A Model Spec update adds an Under-18 Safety Mode.65

25. Feb 2026

    OpenAI releases the Codex desktop app for macOS.29

26. Mar 5, 2026

    OpenAI releases GPT-5.4 and GPT-5.4 Pro with a 1M-token context window.1555

27. Mar 24, 2026

    OpenAI announces the deprecation of its Videos API and Sora 2 models, and the OpenAI Foundation's board chair publishes an update on the Foundation's plans.1868

28. Apr 24, 2026

    OpenAI releases GPT-5.5 and GPT-5.5 Pro in the API.15

29. Jun 26, 2026

    OpenAI begins a limited preview of GPT-5.6 for trusted partners at the U.S. government's request.47

30. Jul 9, 2026

    OpenAI releases GPT-5.6 and merges the Codex app into the ChatGPT desktop app, the same week it launches ChatGPT Work.1529

31. Aug 26, 2026

    The Assistants API shuts down.15

32. Sep 3, 2026

    OpenAI releases GPT-6 Astra, which its system card treats as Critical in cybersecurity.1543

33. Sep 9, 2026

    Paul Christiano joins the OpenAI Foundation board.69

34. Sep 24, 2026

    The listed shutdown date for the Videos API and Sora 2 models arrives.1823

35. Sep 29, 2026

    OpenAI holds DevDay 2026 in San Francisco and releases GPT-6.1 Sol.8115

36. Oct 7, 2026

    OpenAI begins launching GPT-6 Sol and GPT-6 Luna to everyone in ChatGPT.45

37. Oct 8, 2026

    A White House fact sheet lists OpenAI as an industry partner in the Genesis Mission Consortium.75

Leadership

Titles as the official source lists them.

• Sam Altman

  CEO of OpenAI7274

• Sarah Friar

  Chief Financial Officer1

• Bret Taylor

  Chair of the OpenAI Foundation and OpenAI Group PBC Boards7069

• Zico Kolter

  Chair of the Safety and Security Committee692

• Paul Christiano

  Member of the OpenAI Foundation Board and non-voting observer on the OpenAI Group PBC Board69

• David Vélez

  Member of the boards of the OpenAI Foundation and OpenAI Group PBC70

• Robin Vince

  Member of the boards of the OpenAI Foundation and OpenAI Group PBC70

• Wojciech Zaremba

  Head of AI Resilience, OpenAI Foundation68

• Jacob Trefethen

  Head of Life Sciences and Curing Diseases, OpenAI Foundation68

• Anna Makanju

  Head of AI for Civil Society and Philanthropy, OpenAI Foundation68

• Robert Kaiden

  Chief Financial Officer, OpenAI Foundation68