GPT-5 is OpenAI’s fifth-generation flagship AI system, launched on 7 August 2025 as a unified model that pairs a fast, high-throughput model with a deeper reasoning model behind a real-time router. This GPT-5 history timeline traces the full arc of OpenAI’s evolution — from the lab’s founding in December 2015, through GPT-1 (2018), GPT-2, GPT-3, ChatGPT (2022) and GPT-4 (2023), to the GPT-5 generation and its 2026 point releases including GPT-5.5 and GPT-5.6 (Sol, Terra, Luna). Every milestone here is drawn from official OpenAI announcements, research publications and technical documentation, with editorial analysis kept clearly separate from verified fact, and third-party benchmarks labelled as such.
GPT-5 is OpenAI’s fifth flagship generative pre-trained transformer, released on 7 August 2025. Unlike a single model, GPT-5 is a system: a fast model for most queries, a deeper “GPT-5 thinking” reasoning model for hard problems, and a real-time router that picks between them based on the task. OpenAI reports gains over GPT-4 in coding, writing, health answers and factual accuracy, with lower hallucination and stronger agentic tool use. It is available in ChatGPT, Microsoft Copilot and the OpenAI API. The generation has since advanced through GPT-5.1 (Nov 2025), GPT-5.2 (Dec 2025), GPT-5.4 (Mar 2026), GPT-5.5 (Apr 2026) and GPT-5.6 Sol, Terra and Luna (rolling out from July 2026). OpenAI has not announced a GPT-6, and this article avoids speculation about any unreleased model.
Six defining facts about OpenAI, GPT and the road to GPT-5.
Origin
Milestone
Flagship
Latest
Adoption
Models
Reverse chronological — latest developments first, OpenAI’s 2015 founding last.
Official release: After a preview of GPT-5.6 Sol on 26 June 2026, OpenAI began rolling out the GPT-5.6 family from 9 July 2026 across three durable capability tiers — Sol (flagship), Terra (balanced, lower cost) and Luna (fastest, most affordable). Technical innovation: the release adds programmatic tool calling in the Responses API. Developer impact: API pricing (per 1M tokens) is Sol $5 in / $30 out, Terra $2.50 / $15 and Luna $1 / $6. Enterprise impact: OpenAI also shipped ChatGPT Work, an agent on a new desktop app that acts across local files, installed apps and a built-in browser, breaking projects into steps to deliver finished sheets, slides, docs and web apps.
Why it matters: GPT-5.6 decouples model generation from capability tier, letting Sol, Terra and Luna advance on their own cadence — OpenAI’s clearest move yet toward matching model choice to cost and latency. Benchmarks (per OpenAI): GPT-5.6 Sol reports 53.6 on Agents’ Last Exam, 92.2% on BrowseComp and 62.6% on OSWorld 2.0.
Official releases: OpenAI shipped GPT-5.3-Codex (5 February 2026) for agentic coding, GPT-5.4 (5 March 2026), and GPT-5.5 (23 April 2026) in Thinking and Pro variants, with GPT-5.5 Instant becoming ChatGPT’s default model for all users on 5 May 2026. Capability improvements: OpenAI reported GPT-5.5 gains in agentic coding, scientific research and knowledge work while using fewer tokens per task. Benchmarks (per OpenAI): GPT-5.5 posted 82.7% on Terminal-Bench 2.0 and 51.7% on FrontierMath Tier 1–3.
Enterprise impact: around this period OpenAI reported ChatGPT reaching roughly 900 million weekly active users and, by June 2026, about 1 billion monthly active users — figures reported by OpenAI and press, not audited.
Model release: OpenAI unveiled GPT-5 in a livestream on 7 August 2025. Technical innovation: GPT-5 is a system combining a fast, high-throughput model, a deeper “GPT-5 thinking” reasoning model, and a real-time router that decides which to use based on complexity, tool needs and user intent. Capability improvements (per OpenAI): faster responses, better coding and writing, more accurate health answers and lower hallucination, plus agentic abilities such as autonomous browsing. Developer & enterprise impact: GPT-5 shipped simultaneously in ChatGPT, Microsoft Copilot and the OpenAI API. Point releases GPT-5.1 (12 November 2025) and GPT-5.2 (11 December 2025) followed.
Official releases: OpenAI shipped o3-mini (31 January 2025); the research-preview GPT-4.5 “Orion” (27 February 2025), its largest model at the time; the API-focused GPT-4.1 family (14 April 2025) with up to a 1 million-token context window; and the advanced reasoning models o3 and o4-mini (16 April 2025). Developer impact: GPT-4.1 raised coding and long-context performance for API builders, while the o-series brought tool-using “thinking” models to ChatGPT and the API. Why it matters: these releases proved that inference-time reasoning and long context were production-ready — the direct groundwork for GPT-5’s routed design later that year.
Model release: OpenAI launched GPT-4o (“omni”) on 13 May 2024 — a natively multimodal model handling text, audio and images in real time, powering faster voice conversations and a new desktop app. Technical innovation: in September 2024 OpenAI introduced the o1 reasoning model (with o1-mini), trained to “think” step by step before answering, followed by the full o1 and o1 pro in December. Capability improvements: o1 traded speed for deliberation, sharply improving math, science and coding. Developer & enterprise impact: GPT-4o cut API costs and latency, while ChatGPT added Canvas, Search and Projects for structured work.
Model release: OpenAI launched GPT-4 on 14 March 2023 — a multimodal model accepting image and text input, with a major leap in reasoning and professional-exam performance. Technical innovation: 2023 added plugins, browsing and Code Interpreter (Advanced Data Analysis), plus function calling (June 2023) that let models call external tools. Developer impact: at DevDay (6 November 2023) OpenAI shipped GPT-4 Turbo, a 128K context window, custom GPTs and the Assistants API. Enterprise impact: Microsoft’s expanded partnership embedded GPT-4 into Copilot and Azure OpenAI Service.
Product launch: OpenAI released ChatGPT on 30 November 2022, built on GPT-3.5 and refined with reinforcement learning from human feedback (RLHF), following the InstructGPT alignment work published in January 2022. Capability improvements: instruction tuning made the model far easier and safer to converse with than raw GPT-3. Why it matters: ChatGPT reached an estimated 100 million users within two months, the fastest-growing consumer app of its time, and turned generative AI into a global phenomenon. Enterprise impact: ChatGPT Plus launched in February 2023, beginning OpenAI’s consumer-subscription business.
Model release: OpenAI published GPT-3 on 28 May 2020 (the paper “Language Models are Few-Shot Learners”), a 175-billion-parameter model that could perform new tasks from a prompt alone — few-shot in-context learning. Developer impact: the OpenAI API launched on 11 June 2020, making GPT-3 available to developers and creating an entire ecosystem. Technical innovation: in 2021 OpenAI released Codex, fine-tuned on code, which powered GitHub Copilot and brought AI code completion to millions of developers. Why it matters: GPT-3 showed that scaling transformers unlocked general capabilities, setting the template for every model since.
Research & release: OpenAI introduced GPT-1 on 11 June 2018 in the paper “Improving Language Understanding by Generative Pre-Training”, establishing the pre-train then fine-tune recipe on a transformer. Model release: GPT-2 (14 February 2019), with 1.5 billion parameters, generated strikingly fluent text; OpenAI initially withheld the full model over misuse concerns, releasing it in stages through November 2019. Why it matters: GPT-1 proved the generative-pretraining approach and GPT-2 proved it scaled — the two foundations on which GPT-3, ChatGPT and GPT-5 were built. Context: both built on Google’s 2017 transformer architecture (“Attention Is All You Need”).
Founding: OpenAI was announced on 11 December 2015 as a non-profit AI research lab, co-chaired by Sam Altman and Elon Musk, with a founding team including Greg Brockman, Ilya Sutskever, John Schulman and Wojciech Zaremba, and backers pledging up to $1 billion. Mission: to ensure that artificial general intelligence benefits all of humanity. Early research: OpenAI released Gym (2016) for reinforcement-learning research and built game-playing agents such as the OpenAI Five Dota 2 bots. Structure: in 2019 OpenAI created a “capped-profit” entity and secured a $1 billion investment from Microsoft, deepening into a multi-year partnership.
As of 10 July 2026, OpenAI’s latest released flagship generation is GPT-5.6 (Sol, Terra, Luna). OpenAI has not officially announced a “GPT-6” or confirmed the specifications of any future model. In keeping with strict accuracy standards, this article does not present leaks, rumours or analyst speculation about unreleased models as fact. Any such reports should be treated as unconfirmed until OpenAI publishes official details.
Each generation, its headline capability and its main limitation — based on official releases and documentation.
| Model | Released | Headline capability | Main limitation |
|---|---|---|---|
| GPT-1 | Jun 2018 | Proved generative pre-training + fine-tuning on a transformer (117M params) | Small, narrow; needed task-specific fine-tuning |
| GPT-2 | Feb 2019 | Fluent open-ended text; 1.5B params; zero-shot ability | Prone to drift and errors; staged release over misuse fears |
| GPT-3 | May 2020 | 175B params; few-shot in-context learning; powered the API | Hallucinations; no built-in instruction following |
| GPT-3.5 | 2022 | Instruction-tuned with RLHF; the engine behind ChatGPT | Weaker reasoning; limited context; still hallucinates |
| GPT-4 | Mar 2023 | Multimodal (image input); big reasoning and exam gains | Slower and costlier; knowledge cutoff; still errs |
| GPT-4o / o-series | 2024 | Native “omni” multimodality (4o); “thinking” reasoning models (o1–o3) | Reasoning models slower and more expensive per query |
| GPT-4.1 | Apr 2025 | API-focused; up to 1M-token context; strong coding | API-only; superseded within months by GPT-5 |
| GPT-5 | Aug 2025 | Routed system: fast model + reasoning model + real-time router | Router behaviour can be opaque; still not infallible |
| GPT-5.5 / 5.6 | 2026 | Efficiency + agentic gains; tiered Sol/Terra/Luna family | Top tiers costlier; staged, sometimes limited availability |
Parameter counts are given only where officially published (GPT-1 through GPT-3). OpenAI has not disclosed parameter counts for GPT-4 and later; this article does not estimate undisclosed figures.
How the product grew from a chat box into an agentic work platform.
ChatGPT launched on 30 November 2022 as a simple web chat built on GPT-3.5. ChatGPT Plus (February 2023) added GPT-4 access; plugins, browsing and Code Interpreter arrived through 2023, and native iOS and Android apps brought it to mobile. September 2023 added voice and vision; DevDay introduced custom GPTs and, in January 2024, the GPT Store. Through 2024 came Memory, the GPT-4o real-time voice model, a desktop app, Canvas for writing and coding, ChatGPT Search and Projects. In 2025, Deep Research and agentic tools such as Operator turned ChatGPT into something that could act, not just answer. In July 2026, ChatGPT Work extended this into a full agent that runs on the desktop, uses local files, apps and a built-in browser, and delivers finished documents, spreadsheets, slides and shareable web apps — the clearest expression yet of OpenAI’s shift from chatbot to autonomous assistant.
The main products and platforms built around the GPT models.
The shift from a single multimodal model to a routed reasoning system.
| Dimension | GPT-4 (2023) | GPT-5 (2025) |
|---|---|---|
| Architecture | Single multimodal model | System: fast model + reasoning model + router |
| Reasoning | Strong for its time; no explicit “thinking” mode | Built-in deeper reasoning, invoked automatically |
| Model choice | User picks the model | Router selects fast vs thinking per query |
| Hallucination | Noticeable | Reduced (per OpenAI) |
| Agentic use | Added via plugins / function calling | Native, stronger tool use and autonomy |
| Availability | ChatGPT Plus, API, Bing | ChatGPT, Microsoft Copilot, API |
OpenAI’s July 2026 three-tier family — official API pricing per 1M tokens.
| Tier | Positioning | Input / Output ($ per 1M tokens) | Best for |
|---|---|---|---|
| Sol | Flagship — highest intelligence | $5 / $30 | Complex coding, research, agentic workflows |
| Terra | Balanced — performance near GPT-5.5 | $2.50 / $15 | General knowledge work at lower cost |
| Luna | Fastest and most affordable | $1 / $6 | High-volume, latency-sensitive tasks |
A high-level orientation across the leading frontier families. Capabilities and benchmarks vary by version and date — figures below are from published or third-party sources and are not directly comparable across labs.
| Family | Maker | Latest flagship (mid-2026) | Known for |
|---|---|---|---|
| GPT-5 | OpenAI | GPT-5.6 Sol | Routed reasoning, agentic tool use, broad ecosystem (ChatGPT, Copilot, API) |
| Claude | Anthropic | Claude family (e.g. Opus 4.8, Fable 5) | Long-context reasoning, coding, safety focus |
| Gemini | Google DeepMind | Gemini (latest generation) | Native multimodality, Google product and search integration |
| Open-weight | Meta, Mistral, others | Llama / Mistral families | Self-hosting, customisation, on-prem deployment |
Cross-vendor benchmark claims are contested and version-specific. As one published example, OpenAI reported GPT-5.6 Sol scoring 53.6 on “Agents’ Last Exam”, which it said exceeded a competing model by roughly 13 points; such vendor benchmarks should be read alongside independent evaluations. This article does not rank models overall and does not present any single benchmark as definitive.
Reported and estimated figures — attributed and dated; treat as directional, not audited.
| Indicator | Figure | Source / note |
|---|---|---|
| ChatGPT users at ~2 months | ~100M | Early 2023; widely reported |
| Weekly active users | ~900M | Reported by OpenAI / press, early 2026 |
| Monthly active users | ~1 billion | Reported around June 2026 |
| Business customers | 1M+ | Reported by OpenAI, 2026 |
| Developers on the platform | ~4 million | Reported, 2026 |
| GPT model generations | GPT-1 → GPT-5 | Official releases, 2018–2026 |
| GPT-3 parameters | 175 billion | Official (2020); later counts undisclosed |
User and revenue figures are reported by OpenAI or the press and are not independently audited here; they change frequently. Where OpenAI has not disclosed a figure (such as parameter counts for GPT-4 and later), none is invented.
Official release dates for major models, for quick reference.
| Date | Release | Type |
|---|---|---|
| 11 Dec 2015 | OpenAI founded | Organisation |
| 11 Jun 2018 | GPT-1 | Model / research |
| 14 Feb 2019 | GPT-2 | Model |
| 28 May 2020 | GPT-3 | Model |
| 11 Jun 2020 | OpenAI API | Developer platform |
| 30 Nov 2022 | ChatGPT | Product |
| 14 Mar 2023 | GPT-4 | Model |
| 6 Nov 2023 | GPT-4 Turbo, GPTs, Assistants (DevDay) | Platform |
| 13 May 2024 | GPT-4o | Model |
| 12 Sep 2024 | o1 (reasoning) | Model |
| 14 Apr 2025 | GPT-4.1 | Model (API) |
| 16 Apr 2025 | o3 & o4-mini | Model |
| 7 Aug 2025 | GPT-5 | Model / system |
| 23 Apr 2026 | GPT-5.5 | Model |
| 9 Jul 2026 | GPT-5.6 (Sol, Terra, Luna) & ChatGPT Work | Model / product |
Founded in December 2015 to ensure AGI benefits humanity; creator of the GPT models, ChatGPT, the o-series and the OpenAI API, and the driving force behind the modern generative-AI era.
Launched 30 November 2022, OpenAI’s assistant across web, mobile and desktop. Reported to reach roughly a billion monthly users, now extended by the agentic ChatGPT Work.
OpenAI’s fifth flagship generation (7 August 2025): a routed system pairing a fast model with a reasoning model, advanced through GPT-5.1 to GPT-5.6.
OpenAI’s largest partner and investor since 2019, providing Azure compute and embedding OpenAI models across its products through a multi-year, multi-billion-dollar alliance.
Microsoft’s enterprise gateway to OpenAI models, offering GPT-5-generation models with cloud governance, compliance and security for businesses.
The AI pair-programmer built on OpenAI Codex and later models, which brought AI code completion to millions of developers and helped define agentic coding.
Launched June 2020, the programmatic gateway to GPT models. Its Responses API and tool calling let developers build assistants and autonomous agents.
OpenAI co-founder and CEO, a central figure in the company’s founding in 2015 and its strategy through ChatGPT, GPT-4 and the GPT-5 generation.
The most-searched questions about GPT-5 and OpenAI, answered with verified information.
Answers grounded in official OpenAI announcements and documentation, with third-party figures attributed.