The first GPT model came out in 2018. Claude, Gemini, Grok, DeepSeek, and Llama followed in 2023. This page lists major public releases and important named variants from OpenAI, Anthropic, Google, DeepSeek, Meta, and xAI through October 6, 2026.
Scope: this timeline covers major publicly announced model releases and important named variants. It does not list every API snapshot, fine-tune, embedding model, safety model, speech model, or internal research checkpoint. Google, for example, also ships live and text-to-speech variants that are not separate generations.
Jump to [OpenAI](#openai-gpt-and-the-o-series), [Claude](#anthropic-claude), [Gemini](#google-gemini), [DeepSeek](#deepseek), [Llama](#meta-llama), or [Grok](#xai-grok).
Latest major model by company
"Latest" here means the newest major named release from that lab, not the default chat model and not every specialist, live, or voice variant.
| Company | Latest major model | Release | Official source |
|---|---|---|---|
| OpenAI | GPT-6.1 Sol | September 29, 2026 | OpenAI |
| Anthropic | Claude Sonnet 5.5 | September 28, 2026 | Anthropic |
| Gemini 3.8 Flash | September 2, 2026 | ||
| DeepSeek | DeepSeek-V4.1-Flash | September 10, 2026 | DeepSeek |
| Meta | Llama 4 Maverick | April 5, 2025 | Meta |
| xAI | Grok-4.7 | September 21, 2026 | xAI |
GPT-6 Astra is OpenAI's flagship GPT-6 model. GPT-6.1 Sol is a later, lower-cost GPT-6 model focused on agentic coding, computer use, and professional work. OpenAI's help center says GPT-6.1 Sol, GPT-6 Sol, and GPT-6 Luna are for ChatGPT Work, Codex, and the API, not regular ChatGPT conversations. DeepSeek-V4-Pro reached general availability on August 16, 2026. On September 10, DeepSeek made V4.1-Flash the live model and said older Flash names route to it. Use V4.1-Flash when you mean the current DeepSeek model.
The Grok 4.7 news post does not print a calendar date. The September 21 date is the date on xAI's public training summary for that model, and the same day GitHub and Cursor listed it.
The latest-model table has an official link on every row. Older rows are the major public names and dates, and they do not each have a source link yet.
For a coding pick, read best AI for coding in October 2026. For the same tasks run on six models, read the six-model coding test.
OpenAI: GPT and the o-series
OpenAI's lineup has included GPT general-purpose models and the earlier o-series reasoning family. Newer GPT generations combine general use, reasoning, coding, tool use, and agent work in one line. The o-series names below are the older separate reasoning models.
| Model | Release | What changed |
|---|---|---|
| GPT-1 | June 2018 | First GPT generation. |
| GPT-2 | February 14, 2019 | Clear jump in text generation. |
| GPT-3 | June 2020 | 175B model. Few-shot learning became useful. |
| GPT-3.5 | November 30, 2022 | The model behind the first ChatGPT. |
| GPT-4 | March 14, 2023 | Better reasoning, coding, and instruction following. |
| GPT-4 Turbo | November 6, 2023 | Faster, cheaper, 128K context. |
| GPT-4o | May 13, 2024 | Text, vision, and audio in one model. |
| GPT-4o mini | July 18, 2024 | Smaller and cheaper GPT-4o. |
| o1-preview | September 12, 2024 | First public reasoning model. |
| o1-mini | September 12, 2024 | Smaller reasoning model for STEM and code. |
| o1 | December 5, 2024 | Production reasoning model. |
| o3-mini | January 31, 2025 | Cheaper reasoning for code, math, and science. |
| GPT-4.5 | February 27, 2025 | Aimed at conversation, knowledge, and writing. |
| GPT-4.1 | April 14, 2025 | Better coding, instructions, and long context. |
| GPT-4.1 mini | April 14, 2025 | Smaller GPT-4.1. |
| GPT-4.1 nano | April 14, 2025 | Smallest GPT-4.1. |
| o3 | April 16, 2025 | Stronger reasoning, including visual tasks. |
| o4-mini | April 16, 2025 | Faster, cheaper reasoning model. |
| o3-pro | June 10, 2025 | Higher-compute o3. |
| GPT-OSS | August 5, 2025 | Open-weight models in 20B and 120B sizes. |
| GPT-5 | August 7, 2025 | Next main GPT generation. |
| GPT-5.1 | November 12, 2025 | Tighter instruction following. |
| GPT-5.2 | December 11, 2025 | Stronger reasoning and coding. |
| GPT-5.3 Instant | March 3, 2026 | Faster everyday ChatGPT model. |
| GPT-5.3-Codex | 2026 | Coding model for software work and coding agents. |
| GPT-5.4 | March 5, 2026 | Professional work, documents, and agents. |
| GPT-5.4 Pro | March 5, 2026 | Higher-end GPT-5.4. |
| GPT-5.5 | April 23, 2026 | Coding, research, and computer use. |
| GPT-5.5 Pro | April 23, 2026 | Higher-end GPT-5.5. |
| GPT-5.5 Instant | May 5, 2026 | Faster general chat model. |
| GPT-5.6 | July 9, 2026 | Reasoning, coding, cybersecurity, and agents. |
| GPT-5.6 Sol | July 9, 2026 | High-end GPT-5.6 reasoning model. |
| GPT-5.6 Terra | July 9, 2026 | Mid-tier GPT-5.6. |
| GPT-5.6 Luna | July 9, 2026 | Faster, cheaper GPT-5.6. |
| GPT-6 Astra | September 2026 | GPT-6 flagship for computer use, browsing, coding, and science. |
| GPT-6 Sol | September 22, 2026 | Faster, more affordable GPT-6. |
| GPT-6 Luna | September 22, 2026 | Lower-cost GPT-6 for everyday volume. |
| GPT-6.1 Sol | September 29, 2026 | Upgrade to GPT-6 Sol for coding and computer use. |
Anthropic: Claude
Claude uses both generation numbers and tier names. Haiku targets speed and efficiency. Sonnet balances capability and cost. Opus targets the highest-end work. Anthropic also introduced the specialized Fable and Mythos 5.1 models in September 2026.
| Model | Release | What changed |
|---|---|---|
| Claude 1 | March 14, 2023 | First major Claude. |
| Claude 1.3 | May 2023 | Improved Claude 1. |
| Claude Instant | 2023 | Faster, cheaper Claude. |
| Claude 2 | July 11, 2023 | Better coding and long answers. |
| Claude Instant 1.2 | August 9, 2023 | Better math, coding, and reasoning at lower cost. |
| Claude 2.1 | November 21, 2023 | 200K context and better accuracy. |
| Claude 3 family | March 4, 2024 | Haiku, Sonnet, and Opus. |
| Claude 3 Sonnet | March 4, 2024 | Balanced speed and skill. |
| Claude 3 Opus | March 4, 2024 | Top Claude 3 model. |
| Claude 3 Haiku | March 13, 2024 | Fastest Claude 3 model. |
| Claude 3.5 Sonnet | June 20, 2024 | Large jump in coding. |
| Claude 3.5 Haiku | October 22, 2024 | Fast model close to older larger Claudes. |
| Claude 3.7 Sonnet | February 24, 2025 | Optional extended thinking. |
| Claude Opus 4 | May 22, 2025 | Built for long coding and agent jobs. |
| Claude Sonnet 4 | May 22, 2025 | Daily model for coding and agents. |
| Claude Opus 4.1 | August 5, 2025 | Stronger Opus 4. |
| Claude Sonnet 4.5 | September 29, 2025 | Coding, agents, and computer use. |
| Claude Haiku 4.5 | October 2025 | Fast 4.5 model. |
| Claude Opus 4.5 | November 24, 2025 | Coding, agents, and professional work. |
| Claude Opus 4.6 | February 5, 2026 | Long agent runs and large codebases. |
| Claude Sonnet 4.6 | February 17, 2026 | Coding, planning, and computer use. |
| Claude Opus 4.7 | April 16, 2026 | Software engineering and vision. |
| Claude Opus 4.8 | May 28, 2026 | Coding, agents, and professional work. |
| Claude Sonnet 5 | June 30, 2026 | New Sonnet line for agentic coding. |
| Claude Opus 5 | July 24, 2026 | High-end Claude 5. |
| Claude Fable 5.1 | September 1, 2026 | Coding and knowledge work. |
| Claude Mythos 5.1 | September 1, 2026 | Reasoning, cybersecurity, and research. |
| Claude Opus 5.5 | September 22, 2026 | Current high-end Opus. |
| Claude Sonnet 5.5 | September 28, 2026 | Faster Sonnet 5. |
Google: Gemini
Gemini has used Pro models for higher-capability work and Flash models for faster, more efficient work. Google also ships specialist variants for image generation, live interaction, speech, and security. Gemini Omni is a creative multimodal model, not the default coding model.
| Model | Release | What changed |
|---|---|---|
| Gemini 1.0 | December 6, 2023 | First Gemini generation: Nano, Pro, Ultra. |
| Gemini 1.0 Nano | December 2023 | On-device model. |
| Gemini 1.0 Pro | December 2023 | General Gemini 1.0. |
| Gemini 1.0 Ultra | February 2024 | Top Gemini 1.0 model. |
| Gemini 1.5 | February 15, 2024 | Long context and better multimodal input. |
| Gemini 1.5 Pro | February 15, 2024 | High-end long-context model. |
| Gemini 1.5 Flash | May 14, 2024 | Faster Gemini 1.5. |
| Gemini 2.0 | December 11, 2024 | Built for tools and agents. |
| Gemini 2.0 Flash | December 11, 2024 | Fast general 2.0 model. |
| Gemini 2.0 Flash-Lite | February 2025 | Cheaper 2.0 model. |
| Gemini 2.5 | March 25, 2025 | Thinking models for harder problems. |
| Gemini 2.5 Pro | March 25, 2025 | High-end 2.5 reasoning model. |
| Gemini 2.5 Flash | April 2025 | Faster 2.5 model. |
| Gemini 2.5 Flash-Lite | 2025 | Lightweight 2.5 model. |
| Gemini 3 | November 18, 2025 | Reasoning, coding, and agents. |
| Gemini 3 Pro | November 18, 2025 | High-end Gemini 3. |
| Gemini 3 Flash | December 2025 | Faster Gemini 3. |
| Gemini 3.1 Pro | February 19, 2026 | Better reasoning and coding. |
| Gemini 3.1 Flash-Lite | 2026 | Cheap high-volume Flash. |
| Gemini 3.5 Flash | May 19, 2026 | Fast model for agent work. |
| Gemini Omni | May 19, 2026 | Turns text, image, and video into generated video and other creative output. |
| Gemini 3.5 Flash-Lite | July 21, 2026 | Low-latency agent model. |
| Gemini 3.5 Flash Cyber | July 21, 2026 | Security-focused 3.5 Flash. |
| Gemini 3.6 Flash | July 21, 2026 | Better coding and multimodal work. |
| Gemini 3.7 Flash | August 13, 2026 | Coding and agents. |
| Gemini 3.8 Flash | September 2, 2026 | Latest main Flash model. |
| Gemini 3.8 Flash Cyber | September 2, 2026 | Security variant of 3.8 Flash. |
DeepSeek
DeepSeek uses three names. V models are general. Coder models are for code. R1 is the reasoning line. Later V models fold reasoning and tools into the main line.
| Model | Release | What changed |
|---|---|---|
| DeepSeek-LLM | November 2023 | First general DeepSeek family. |
| DeepSeek-Coder | November 2023 | First coding family. |
| DeepSeek-V2 | May 17, 2024 | Mixture-of-experts, long context. |
| DeepSeek-Coder-V2 | June 14, 2024 | Much stronger code model. |
| DeepSeek-V2-0628 | June 28, 2024 | Better code, math, and reasoning. |
| DeepSeek-Coder-V2-0724 | July 24, 2024 | Updated Coder-V2. |
| DeepSeek-V2.5 | September 5, 2024 | General and coding in one model. |
| DeepSeek-V2.5-1210 | December 10, 2024 | Better math, code, and writing. |
| DeepSeek-R1-Lite-Preview | November 20, 2024 | Early long-chain reasoning preview. |
| DeepSeek-V3 | December 26, 2024 | Large open mixture-of-experts model. |
| DeepSeek-R1 | January 20, 2025 | Main reasoning model. |
| DeepSeek-R1-Distill | January 20, 2025 | Smaller models trained from R1. |
| DeepSeek-V3-0324 | March 24, 2025 | Better reasoning, front-end code, and tools. |
| DeepSeek-R1-0528 | May 28, 2025 | Updated R1. |
| DeepSeek-V3.1 | August 21, 2025 | Hybrid thinking and non-thinking. |
| DeepSeek-V3.1-Terminus | September 22, 2025 | Language consistency and agents. |
| DeepSeek-V3.2-Exp | September 29, 2025 | Experimental V3.2. |
| DeepSeek-V3.2 | December 1, 2025 | Reasoning-first, built for tools. |
| DeepSeek-V3.2-Speciale | December 1, 2025 | Higher-reasoning V3.2. |
| DeepSeek-V4 Preview | April 24, 2026 | 1M-token context and stronger agents. |
| DeepSeek-V4-Pro | April 24, 2026 | High-end V4. |
| DeepSeek-V4-Flash | April 24, 2026 | Faster, smaller V4. |
| DeepSeek-V4-Flash public beta | July 31, 2026 | Updated Flash with better coding. |
| DeepSeek-V4-Pro GA | August 16, 2026 | Production V4 Pro. |
| DeepSeek-V4-Flash-Vision-Exp | August 21, 2026 | Experimental model with visual input. |
| DeepSeek-V4.1-Flash | September 10, 2026 | Faster multimodal Flash, lower cost. |
Meta: Llama
Llama is the open-weight family. This timeline stops at Llama 4 in April 2025. There is no 2026 Llama release in this list.
| Model | Release | What changed |
|---|---|---|
| LLaMA | February 24, 2023 | First LLaMA generation. |
| Llama 2 | July 18, 2023 | Research and commercial use. |
| Llama 2-Chat | July 18, 2023 | Chat-tuned Llama 2. |
| Llama 3 | April 18, 2024 | Better language, reasoning, and code. |
| Llama 3 8B | April 18, 2024 | Smaller Llama 3. |
| Llama 3 70B | April 18, 2024 | Larger Llama 3. |
| Llama 3.1 | July 23, 2024 | 128K context. |
| Llama 3.1 405B | July 23, 2024 | Flagship open model. |
| Llama 3.2 | September 25, 2024 | Vision models and small edge models. |
| Llama 3.2 1B | September 25, 2024 | Tiny text model. |
| Llama 3.2 3B | September 25, 2024 | Small text model. |
| Llama 3.2 11B Vision | September 25, 2024 | Smaller vision model. |
| Llama 3.2 90B Vision | September 25, 2024 | Larger vision model. |
| Llama 3.3 | December 2024 | 70B text model, cheaper to serve. |
| Llama 4 | April 5, 2025 | Multimodal mixture-of-experts generation. |
| Llama 4 Scout | April 5, 2025 | Efficient model, very long context. |
| Llama 4 Maverick | April 5, 2025 | Larger reasoning, code, and vision model. |
| Llama 4 Behemoth | April 5, 2025 | Teacher model. Not a normal public release. |
xAI: Grok
Grok started as a chat model tied to live posts on X. Later versions add vision, longer context, and long coding jobs.
| Model | Release | What changed |
|---|---|---|
| Grok | November 3, 2023 | First Grok chat model. |
| Grok-1 | November 2023 | The original model behind that chat. |
| Grok-1.5 | March 28, 2024 | Better reasoning and code, 128K context. |
| Grok-1.5V | April 12, 2024 | First Grok that could read images. |
| Grok-2 | August 13, 2024 | Better chat, code, and reasoning. |
| Grok-2 Mini | August 13, 2024 | Smaller Grok-2. |
| Grok-3 | February 19, 2025 | Major reasoning model. |
| Grok-3 Mini | February 19, 2025 | Smaller Grok-3. |
| Grok-3 Reasoning | 2025 | Reasoning mode of Grok-3. |
| Grok-4 | July 9, 2025 | Tools and live search. |
| Grok-4 Heavy | July 9, 2025 | Higher-compute Grok-4. |
| Grok-4.1 | November 17, 2025 | Better everyday use. |
| Grok-4.1 Fast | November 2025 | Faster agent and API model. |
| Grok-4.3 | June 17, 2026 | 1M context, adjustable reasoning. |
| Grok-4.5 | July 16, 2026 | Coding and knowledge work. |
| Grok-4.6 | August 12, 2026 | Long agent runs and visual work. |
| Grok-4.7 | September 21, 2026 | Latest Grok for long coding jobs. |
The short generation path
These arrows skip mini, Flash, and Pro variants. Use the tables above when you need a specific date.
OpenAI: GPT-1, GPT-2, GPT-3, GPT-3.5, GPT-4, GPT-4 Turbo, GPT-4o, o1, o3, o4-mini, GPT-4.1, GPT-5, GPT-5.1 through GPT-5.6, GPT-6 Astra, GPT-6 Sol and Luna, GPT-6.1 Sol.
Anthropic: Claude 1, Claude 2, Claude 2.1, Claude 3, Claude 3.5, Claude 3.7, Claude 4, then 4.1, 4.5, 4.6, 4.7, 4.8, Claude 5, Claude 5.1, Claude 5.5.
Google: Gemini 1.0, 1.5, 2.0, 2.5, 3, 3.1, 3.5, 3.6, 3.7, 3.8.
DeepSeek: DeepSeek-LLM, Coder, V2, Coder-V2, V2.5, V3, R1, V3.1, V3.2, V4, V4.1.
Meta: LLaMA, Llama 2, Llama 3, 3.1, 3.2, 3.3, Llama 4.
xAI: Grok, Grok-1.5, Grok-1.5V, Grok-2, Grok-3, Grok-4, Grok-4.1, Grok-4.3, Grok-4.5, Grok-4.6, Grok-4.7.
FAQ
What was the first GPT model?
GPT-1, in June 2018.
When was GPT-4 released?
March 14, 2023.
When was Claude first released?
Claude 1, on March 14, 2023.
When was Gemini released?
Gemini 1.0, on December 6, 2023.
When was DeepSeek first released?
DeepSeek-LLM and DeepSeek-Coder, in November 2023.
When was Llama first released?
The original LLaMA, on February 24, 2023.
When was Grok first released?
November 3, 2023.
What is the latest model in October 2026?
It depends on the lab. Use the latest-model table at the top. OpenAI's newest listed model is GPT-6.1 Sol. DeepSeek's current live model is V4.1-Flash. Meta's newest listed model is still Llama 4 from April 2025.
How to use this page
If you only need a date, use the table for that lab. A Flash, Luna, Haiku, or mini model is the cheaper daily model. A Pro, Opus, Astra, or Sol model is the one to try when a wrong edit is expensive.
The coding pick, with the limit that the test scores are not in yet, is in best AI for coding in October 2026.