All vikis
007 First Light
AI
Aion 2
Ananta
ARC Raiders
ArcheAge Chronicles
Call of Duty: Modern Warfare 4
Chrono Odyssey
Civilization 3
Civilization 4
Claude
Crimson Desert
Deadlock
Diablo 4
Diablo 5
Enshrouded
Fable
Final Fantasy Resonance
Football Manager 27
Forza Horizon 6
God of War
Grand Theft Auto 6
Heroes of Might and Magic 3GPT Model History
OpenAI's GPT models run from GPT-1 (Jun 11, 2018) through GPT-2 (1.5 billion parameters, 2019), GPT-3 (175 billion, 2020), GPT-4 (2023), GPT-4o (2024), GPT-5 (Aug 7, 2025) and the GPT-5.x updates to GPT-6 Astra (Sep 3, 2026) and its siblings.
OpenAI's GPT models began with the work published on Jun 11, 2018 that is now called GPT-1, and reached GPT-6 Astra on Sep 3, 2026. OpenAI published the sizes of GPT-2 (1.5 billion parameters) and GPT-3 (175 billion); none of the sources here gives a size for GPT-4 or any later flagship, and the only later counts are for the open-weight gpt-oss models. The table below lists every major GPT model with its date and what it changed; reasoning models (o1, o3, o4-mini) are included because ChatGPT used them beside GPT models until GPT-5 merged the two lines.
Quick facts
| First GPT | Jun 11, 2018 ("Improving language understanding with unsupervised learning") |
| Largest published size | 175 billion parameters, GPT-3 (2020) |
| First in ChatGPT | A GPT-3.5 series model, Nov 30, 2022 |
| Newest | GPT-6.1 Terra, Oct 26, 2026 (flagship: GPT-6 Astra, Sep 3, 2026) |
| ChatGPT default | GPT-6 (Sol on paid plans, Luna on Free and Go), as of Nov 6, 2026 |
Every GPT model, 2018 to 2026
"Not given" means the sources cited do not state a size.
| Model | Released | Size | What it did |
|---|---|---|---|
| GPT-1 | Jun 11, 2018 | Not given | Showed a Transformer pretrained on unlabeled text and then fine-tuned could set new results on reading and commonsense tests |
| GPT-2 | Feb 14, 2019 | 1.5 billion | Trained on 40GB of text from 8 million web pages; released in stages over misuse concerns, in full on Nov 5, 2019 |
| GPT-3 | May 28, 2020 (paper) | 175 billion | Few-shot learning from examples in the prompt; sold through the OpenAI API from Jun 11, 2020 |
| Codex | Aug 10, 2021 | Not given | A GPT-3 descendant trained on code that powered GitHub Copilot; retired in March 2023 |
| InstructGPT | Mar 4, 2022 (paper) | 1.3 billion (preferred by raters over 175 billion GPT-3) | Followed instructions after training with human feedback |
| GPT-3.5 | Early 2022 (finished training) | Not given | The base of the first ChatGPT, Nov 30, 2022 |
| GPT-4 | Mar 14, 2023 | Not given | Took images as well as text; about the top 10% on a simulated bar exam |
| GPT-4o | May 13, 2024 | Not given | "Omni": text, audio and images in one model; GPT-4-class model on ChatGPT Free |
| GPT-4o mini | Jul 18, 2024 | Not given | Small, cheap model that "surpasses GPT-3.5 Turbo" |
| o1-preview, o1-mini | Sep 12, 2024 | Not given | First reasoning models, thinking before answering |
| o3-mini | Jan 31, 2025 | Not given | Small reasoning model, also on Free |
| GPT-4.5 | Feb 27, 2025 | Not given | Research preview, "our largest" chat model at the time |
| o3, o4-mini | Apr 16, 2025 | Not given | Reasoning models; o3-pro followed for Pro on Jun 10, 2025 |
| GPT-4.1 | May 14, 2025 (in ChatGPT) | Not given | Added to ChatGPT for Plus, Pro and Team under "more models" |
| gpt-oss | Aug 5, 2025 | 117 billion and 21 billion | Open-weight reasoning models under Apache 2.0 |
| GPT-5 | Aug 7, 2025 | Not given | One system that routes between a fast model and "GPT-5 thinking"; replaced GPT-4o, o3 and GPT-4.5 in ChatGPT |
| GPT-5.1 | Nov 12, 2025 | Not given | Instant and Thinking versions |
| GPT-5.2 | Dec 11, 2025 | Not given | Instant, Thinking and Pro; xhigh reasoning in the API |
| GPT-5.3 Instant | Mar 3, 2026 | Not given | Update to ChatGPT's most-used model |
| GPT-5.4 | Mar 5, 2026 | Not given | Thinking and Pro; native computer use |
| GPT-5.5 | Apr 23, 2026 | Not given | New flagship; GPT-5.5 Instant became the default on May 5, 2026 |
| GPT-5.6 Sol, Terra, Luna | Jul 9, 2026 | Not given | First family named by size: flagship, balanced, most cost-efficient |
| GPT-6 Astra | Sep 3, 2026 | Not given | Flagship; OpenAI's first model rated "Critical" for cybersecurity |
| GPT-6 Sol, GPT-6 Luna | Sep 22, 2026 | Not given | Cheaper GPT-6 models; GPT-6 in ChatGPT's Chat tab from Oct 7, 2026 |
| GPT-6.1 Sol | Sep 29, 2026 | Not given | Near-Astra results at a fifth of Astra's token prices, per OpenAI |
| GPT-6.1 Terra | Oct 26, 2026 | Not given | Fast model for visual reasoning, computer use and writing |
The current lineup, with prices and context windows, is on ChatGPT Models and GPT-6.
From research papers to products (2018 to 2022)
GPT-1 was a research result, not a product. GPT-2 is remembered for its release: OpenAI held back the full model at first and published it in stages, the 1.5 billion parameter version last (OpenAI). GPT-3 was the first GPT sold to developers, through the API that opened on Jun 11, 2020. InstructGPT showed that a model trained with human feedback could beat a model more than 100 times its size in raters' eyes, and ChatGPT, a sibling of InstructGPT, used that method on a GPT-3.5 model (OpenAI).
GPT-4 to GPT-4o and the reasoning models (2023 to 2025)
GPT-4 reached ChatGPT Plus on Mar 14, 2023. GPT-4o, in May 2024, made a GPT-4-level model free and handled voice natively. In September 2024 OpenAI started a second line, the o-series OpenAI Reasoning Models, which spend time thinking before they answer; o1, o3 and o4-mini sat beside GPT-4o in ChatGPT's model picker until GPT-5. GPT-4o's retirement from ChatGPT on Feb 13, 2026 came after a reversal: OpenAI had removed it at the GPT-5 launch and restored it after feedback from Plus and Pro users.
GPT-5 and the point releases (2025 and 2026)
GPT-5 joined the two lines into one system with a router. OpenAI then shipped point releases every few months: 5.1, 5.2, 5.3, 5.4, 5.5 and 5.6, the last of them in three named sizes. It says models "generally remain available in ChatGPT for 90 days after a successor is released" (OpenAI Help Center), which is why the list of Retired ChatGPT Models grows quickly. The API keeps models longer: GPT-4o, GPT-4o mini and GPT-4.1 shut down there on Jan 14, 2027.
GPT-6 (2026)
GPT-6 kept the named sizes: Astra on top, Sol in the middle, Luna as the fast, cheap model, then 6.1 updates of Sol and Terra. OpenAI's developer video introduced Astra as "our latest frontier model". Their benchmark results, OpenAI's and independent ones, are on OpenAI Benchmarks.
Frequently asked questions
When was GPT-1 released? OpenAI published the work now called GPT-1 on Jun 11, 2018, in a post titled "Improving language understanding with unsupervised learning".
How many parameters does GPT-4 have? OpenAI has not stated GPT-4's size in any source cited here. The last flagship GPT whose size OpenAI published was GPT-3, at 175 billion parameters; GPT-2 had 1.5 billion.
What GPT model does ChatGPT use now? As of Nov 6, 2026, ChatGPT's Chat tab uses GPT-6: GPT-6 Sol on paid plans and GPT-6 Luna on Free and Go, with GPT-6 Astra behind the Pro option.
What came between GPT-4 and GPT-5? Between GPT-4 (Mar 14, 2023) and GPT-5 (Aug 7, 2025), OpenAI released GPT-4o, GPT-4o mini, GPT-4.5, GPT-4.1 and the reasoning models o1, o3-mini, o3 and o4-mini.
Was there a GPT-3.5? Yes. The first ChatGPT, on Nov 30, 2022, was fine-tuned from a model in the GPT-3.5 series that finished training in early 2022.
Which GPT models are open-weight? OpenAI's open-weight models are gpt-oss-120b (117 billion parameters) and gpt-oss-20b (21 billion), released Aug 5, 2025 under the Apache 2.0 license.
Sources
- Improving language understanding with unsupervised learning, OpenAI, Jun 11, 2018
- Better language models and their implications, OpenAI, Feb 14, 2019
- GPT-2: 1.5B release, OpenAI, Nov 5, 2019
- Language Models are Few-Shot Learners, arXiv, May 28, 2020
- OpenAI API, OpenAI, Jun 11, 2020
- OpenAI Codex, OpenAI, Aug 10, 2021
- Training language models to follow instructions with human feedback, arXiv, Mar 4, 2022
- Introducing ChatGPT, OpenAI, Nov 30, 2022
- GPT-4, OpenAI, Mar 14, 2023
- GPT-4 Technical Report, arXiv, Mar 15, 2023
- Hello GPT-4o, OpenAI, May 13, 2024
- Introducing OpenAI o1, OpenAI, Sep 12, 2024
- Introducing GPT-4.5, OpenAI, Feb 27, 2025
- Introducing OpenAI o3 and o4-mini, OpenAI, Apr 16, 2025
- Welcome OpenAI's gpt-oss, Hugging Face, Aug 5, 2025
- Introducing GPT-5, OpenAI, Aug 7, 2025
- GPT-5.1, OpenAI, Nov 12, 2025
- Introducing GPT-5.2, OpenAI, Dec 11, 2025
- Retiring GPT-4o and older models, OpenAI, Jan 29, 2026
- GPT-5.3 Instant, OpenAI, Mar 3, 2026
- Introducing GPT-5.4, OpenAI, Mar 5, 2026
- Introducing GPT-5.5, OpenAI, Apr 23, 2026
- GPT-5.5 Instant, OpenAI, May 5, 2026
- GPT-5.6, OpenAI, Jul 9, 2026
- GPT-6 Astra, OpenAI, Sep 3, 2026
- Introducing GPT-6 Sol and Luna, OpenAI, Sep 22, 2026
- GPT-6 and Intelligent UI for everyone, OpenAI, Oct 7, 2026
- OpenAI unveils GPT-6.1 Sol at DevDay, Unite.AI, Sep 29, 2026
- API changelog, OpenAI, read Nov 6, 2026
- Deprecations, OpenAI, read Nov 6, 2026
- Model Release Notes, OpenAI Help Center, read Nov 6, 2026
- ChatGPT release notes, OpenAI Help Center, read Nov 6, 2026
- Retiring GPT-4o and other ChatGPT models, OpenAI Help Center, read Nov 6, 2026
- Meet GPT-5.6, OpenAI, Jul 9, 2026 (video)
- Introducing GPT-6 Astra for developers, OpenAI, Sep 4, 2026 (video)
Moments
Meet GPT-5.6
OpenAI
Introducing GPT-6 Astra for developers
OpenAI
Related
- CanvasCanvas was a ChatGPT workspace for editing writing and code beside the chat, launched in beta on Oct 3, 2024. It was removed from current models on May 28, 2026; writing blocks and interactive code blocks in the chat replaced it.
-
-
ChatGPT PulseChatGPT Pulse gave Pro users a daily set of personalized updates that ChatGPT researched overnight, from Sep 25, 2025. OpenAI announced its sunset on Jun 17, 2026, gave Pro users 14 more days, and moved proactive updates into scheduled tasks.
- DALL-EDALL-E was OpenAI's text-to-image model line: DALL-E (Jan 2021), DALL-E 2 (2022) and DALL-E 3 (2023, built into ChatGPT). GPT-4o image generation replaced it in ChatGPT in March 2025, the API models shut down on May 12, 2026, and the DALL-E GPT retired on Aug 30, 2026.
- Group ChatsChatGPT group chats let several people and ChatGPT share one conversation. OpenAI piloted them on Nov 13, 2025, opened them to all logged-in Free, Go, Plus and Pro users on Nov 20, 2025, and stopped new group chats from Jul 9, 2026.
-
History of ChatGPTChatGPT came out on Nov 30, 2022, as a free research preview from OpenAI built on a GPT-3.5 model. By Oct 2026 it had more than 1.2 billion weekly users, paid plans from $8 to $500 a month and the GPT-6 models.
- History of OpenAIOpenAI was founded as a nonprofit AI research company on Dec 11, 2015, added a capped-profit arm in 2019, partnered with Microsoft, released ChatGPT in 2022 and became OpenAI Group PBC under the OpenAI Foundation on Oct 28, 2025.
- OpenAIOpenAI is the AI company behind ChatGPT, Codex and the GPT models. Since Oct 28, 2025 it has been OpenAI Group PBC, controlled by the nonprofit OpenAI Foundation, with Microsoft holding about 27%; CEO Sam Altman leads it, and it was last valued at $852 billion.
See also
-
- gpt-ossgpt-oss is OpenAI's family of open-weight models: gpt-oss-120b and gpt-oss-20b, released Aug 5, 2025 under the Apache 2.0 license, which run on one 80 GB GPU or a 16 GB machine. As of Nov 6, 2026 OpenAI has released no newer general open-weight model.
-
GPT-4oGPT-4o was OpenAI's "omni" model, launched May 13, 2024, that brought native voice and image generation to ChatGPT. It was retired from ChatGPT on Feb 13, 2026 (Apr 3, 2026 in business custom GPTs) and shuts down in the API on Jan 14, 2027.
-
- GPT-6 LunaGPT-6 Luna is OpenAI's cheapest and fastest GPT-6 model, released Sep 22, 2026 at $0.10 per million input tokens and $0.50 output. OpenAI's pricing page and Oct 7 release note name it as ChatGPT's model for Free and Go, though its help articles say GPT-5.6 Luna.
-