GPT-1
A Transformer using generative pretraining followed by task-specific fine-tuning, starting the GPT line.
FAMILY / OPENAI
GPT
The GPT family, from generative pretraining to multimodality, reasoning, and agent systems.
- First model
- 2018
- Models
- 41
- Latest dated version
- GPT-Live-1.1 Quip
01 / RELEASE SEQUENCE
Family release sequence
- 2018-06-11GPT-1Language
- 2019-02-14GPT-2Language
- 2020-05-28GPT-3Language
- 2022-01-27InstructGPTLanguage
- 2022-11-30ChatGPT (GPT-3.5)Language
- 2023-03-01GPT-3.5 TurboLanguage
- 2023-03-14GPT-4Language · Reasoning
- 2023-11-06GPT-4 TurboLanguage
- 2024-05-13GPT-4oLanguage · Multimodal
- 2024-07-18GPT-4o miniLanguage · Multimodal
- 2025-02-27GPT-4.5 PreviewLanguage · Multimodal
- 2025-04-14GPT-4.1Language · Coding
- 2025-04-14GPT-4.1 miniLanguage · Multimodal
- 2025-04-14GPT-4.1 nanoLanguage · Multimodal
- 2025-08-07GPT-5Language · Reasoning
- 2025-08-07GPT-5 miniLanguage · Reasoning
- 2025-08-07GPT-5 nanoLanguage · Reasoning
- 2025-11-13GPT-5.1Language · Reasoning
- 2025-12-11GPT-5.2Language · Reasoning
- 2025-12-18GPT-5.2-CodexCoding · Agent
- 2026-02-05GPT-5.3-CodexCoding · Agent
- 2026-02-12GPT-5.3-Codex-SparkCoding · Agent
- 2026-03-03GPT-5.3 InstantLanguage · Multimodal
- 2026-03-05GPT-5.4Language · Reasoning
- 2026-03-17GPT-5.4 miniReasoning · Coding
- 2026-03-17GPT-5.4 nanoLanguage · Coding
- 2026-04-14GPT-5.4-CyberReasoning · Coding
- 2026-04-23GPT-5.5Language · Reasoning
- 2026-05-05GPT-5.5 InstantLanguage · Multimodal
- 2026-06-26GPT-5.6Language · Reasoning
- 2026-08-10GPT-5.6 CyberReasoning · Coding
- 2026-09-03GPT-6 AstraLanguage · Reasoning
- 2026-09-10GPT-Live-1Language · Agent
- 2026-09-11GPT-RosalindLanguage · Reasoning
- 2026-09-22GPT-6 LunaLanguage · Reasoning
- 2026-09-22GPT-6 SolLanguage · Reasoning
- 2026-09-29GPT-6.1 SolLanguage · Reasoning
- 2026-10-07GPT-6 Luna (October)Language · Reasoning
- 2026-10-07GPT-6 Sol (October)Language · Reasoning
- 2026-10-15GPT-Live-1.1Language · Multimodal
- 2026-10-15GPT-Live-1.1 QuipLanguage · Multimodal
02 / GENERATIONS
What changed
GPT-2
GPT-2 brought the scale-driven emergence of general language behavior into public view—and made release governance part of the model story.
GPT-3
GPT-3 turned prompts into a general-purpose task interface and catalyzed the commercial model API ecosystem.
InstructGPT
GPT-3 models aligned with human feedback to follow instructions; this node represents a series, not a single parameter tier.
ChatGPT (GPT-3.5)
ChatGPT’s original dialogue model, fine-tuned from GPT-3.5 with human feedback and introduced as a free research preview.
GPT-3.5 Turbo
The GPT-3.5 model for the message-based chat API, initially released as the gpt-3.5-turbo-0301 snapshot.
GPT-4
GPT-4 combined strong language reasoning and image understanding in one product generation and became a base for early tool-using agents.
GPT-4 Turbo
Entered developer preview with a 128K context window, expanding GPT-4’s long-document capabilities.
GPT-4o
GPT-4o moved native multimodality from a research capability into a real-time mass-market experience.
GPT-4o mini
A text-and-vision model for inexpensive, low-latency calls, launched with 128K context and text output.
GPT-4.5 Preview
A research preview scaling unsupervised learning for knowledge, natural conversation, and creative writing rather than test-time reasoning.
GPT-4.1
Improved coding, instruction following, and long-context processing, with up to one million tokens in the API.
GPT-4.1 mini
The smaller multimodal GPT-4.1 tier balancing speed, cost, and long context.
GPT-4.1 nano
The lowest-latency GPT-4.1 tier for high-volume tasks such as classification and completion.
GPT-5
It folded reasoning into the default product experience and established agentic coding as a core GPT capability.
GPT-5 mini
The lower-cost GPT-5 tier for workloads balancing reasoning quality, throughput, and price.
GPT-5 nano
The smallest GPT-5 tier for classification, extraction, ranking, and lightweight subtasks.
GPT-5.1
An adaptive-reasoning API update that reduced latency on easy tasks and added apply_patch and shell tools.
GPT-5.2
A GPT-5 series upgrade for professional work and long-running agents, offered in Instant, Thinking, and Pro tiers.
GPT-5.2-Codex
A GPT-5.2 variant optimized for long-horizon software engineering, large code changes, Windows, and context compaction.
GPT-5.3-Codex
A Codex model extending agentic coding into research, tool use, and end-to-end computer work.
GPT-5.3-Codex-Spark
A small, ultra-fast Codex research preview designed for real-time interactive coding.
GPT-5.3 Instant
A GPT-5.3 update for everyday conversation, focused on fewer unnecessary refusals, better search accuracy, and smoother writing.
GPT-5.4
Unified GPT-5.3-Codex coding advances with general reasoning, tools, and professional document work.
GPT-5.4 mini
A fast small model for coding and subagents with text, image, tools, and a 400K context window.
GPT-5.4 nano
The smallest GPT-5.4 tier for high-throughput extraction, ranking, and simple coding subtasks.
GPT-5.4-Cyber
A GPT-5.4 cybersecurity model for vetted defenders, with a lower refusal boundary for legitimate vulnerability research and binary reverse engineering; scheduled for retirement under the official notice.
GPT-5.5
A 1M-context frontier model for complex knowledge work, computer use, coding, and scientific research.
GPT-5.5 Instant
The everyday fast tier of GPT-5.5, improving accuracy, concision, and use of personalized context.
GPT-5.6
This generation separated model versioning from durable capability tiers and incorporated multi-agent coordination at its highest effort level.
GPT-5.6 Cyber
A purpose-trained cybersecurity model for authorized vulnerability research, exploit validation, and security testing, improving selected defensive workflows.
GPT-6 Astra
Astra advances the GPT line toward longer end-to-end workflows with a 1.05M context window, 128K output limit, and stronger tool use as the new flagship specification.
GPT-Live-1
GPT-Live-1 turns real-time voice from a chained speech-recognition, language-model, and speech-synthesis pipeline into a unified API voice layer while keeping the backend model replaceable.
GPT-Rosalind
GPT-Rosalind extends GPT capabilities into eligibility-reviewed life-sciences research, connecting model intelligence with scientific tools and reviewable workflows.
GPT-6 Luna
Luna extends the GPT-6 line to high-volume workloads, forming a clear current API, Codex, and Work tier alongside Astra and Sol.
GPT-6 Sol
Sol joins Astra and Luna as a distinct GPT-6 tier, bringing frontier capability to the API, Codex, and Work at lower cost.
GPT-6.1 Sol
GPT-6.1 Sol was introduced in OpenAI’s system card as a new model family in the GPT-6 series; its API page lists 1.05M context and 128K output, with Responses API tools and Multi-agent Beta.
GPT-6 Luna (October)
The October version introduces Intelligent UI: GPT-6 can combine text, visuals, and interactive components, and begin answering while it thinks or uses tools; OpenAI also updated safety training.
GPT-6 Sol (October)
The October version introduces Intelligent UI: GPT-6 can combine text, visuals, and interactive components, and begin answering while it thinks or uses tools; OpenAI also updated safety training.
GPT-Live-1.1
OpenAI’s system card identifies it as the successor to GPT-Live-1; it is recorded as a model version, without treating the ChatGPT Voice rollout as an API release.
GPT-Live-1.1 Quip
OpenAI explicitly identifies Quip as the successor to GPT-Live-1 mini and as the mini version of GPT-Live-1.1; the older mini entity is not backfilled outside this task window.