GPT-1

Jun 18·117M

text

Research only

Improving Language Understanding by Generative Pre-Training — established the GPT paradigm.

→ 1 successor

Open details →

GPT-2

Feb 19·1.5B

text

Open weight

Larger GPT with staged release after concerns about misuse; demonstrated zero-shot task transfer.

→ 1 successor

Open details →

GPT-3

Jun 20·175B

text

Commercial API

Few-shot learners at scale — catalyzed the commercial LLM era via API access.

→ 1 successor

Open details →

PaLM

Apr 22·540B

text

Research only

Google Pathways Language Model demonstrating few-shot scaling at 540B.

→ 1 successor

Open details →

GPT-3.5

Nov 22·~175B class

text

Commercial API

Instruction-tuned GPT-3 lineage powering early ChatGPT (text-davinci-003 / gpt-3.5-turbo).

→ 1 successor

Open details →

Stable Diffusion 1.0

Aug 22·~1B

vision

Open weight

Open latent diffusion model that democratized high-quality image generation.

→ 1 successor

Open details →

Claude 1

Mar 23·Undisclosed

text

Commercial API

First public Claude assistant from Anthropic, trained with Constitutional AI ideas.

→ 1 successor

Open details →

Claude 2

Jul 23·Undisclosed

text

Commercial API

Longer context Claude with improved coding and reasoning; claude.ai consumer product.

→ 1 successor

Open details →

PaLM 2

May 23·Undisclosed

text

Commercial API

Powered early Bard / Vertex AI; improved multilingual and reasoning over PaLM.

Open details →

Gemini 1.0

Dec 23·Undisclosed

multimodal

Commercial API

First Gemini generation natively multimodal across Ultra, Pro, and Nano tiers.

→ 1 successor

Open details →

LLaMA

Feb 23·7B–65B

text

Research only

Meta research release that unlocked a wave of open fine-tunes (Alpaca, Vicuna, etc.).

→ 1 successor

Open details →

Llama 2

Jul 23·7B–70B

text

Open weight

Open-weight commercial-friendly Llama with chat fine-tunes; industry standard base for 2023–24.

→ 1 successor

Open details →

Mistral 7B

Sep 23·7B

text

Open source

Apache-licensed 7B that punched above its size and put European open models on the map.

→ 1 successor

Open details →

Mixtral 8x7B

Dec 23·46.7B (12.9B active)

text

Open source

Sparse MoE open model competitive with much larger dense models.

→ 1 successor

Open details →

Falcon 40B

May 23·40B

text

Open source

UAE open model that briefly topped open LLM leaderboards in 2023.

→ 1 successor

Open details →

Falcon 180B

Sep 23·180B

text

Open weight

Large open-weight Falcon trained on RefinedWeb.

Open details →

GPT-4

Mar 23·Undisclosed

multimodal

Commercial API

Flagship multimodal model with strong reasoning and exam performance; vision via GPT-4V.

→ 2 successors

Open details →

Qwen2

Jun 24·0.5B–72B

text

Open weight

Strong multilingual open Qwen generation widely used in Asia and open ecosystems.

→ 1 successor

Open details →

Qwen2.5

Sep 24·0.5B–72B

text

Open weight

Improved coding and instruction-following Qwen generation.

→ 1 successor

Open details →

Claude 3

Mar 24·Undisclosed

multimodal

Commercial API

Claude 3 family (Opus, Sonnet, Haiku) with vision and strong reasoning.

→ 1 successor

Open details →

Claude 3.5 Sonnet

Jun 24·Undisclosed

multimodal

Commercial API

Mid-2024 Claude upgrade that set a new bar for coding and computer use demos.

→ 1 successor

Open details →

Command R

Mar 24·35B

text

Open weight

RAG-optimized enterprise model from Cohere with tool use and long context.

→ 1 successor

Open details →

Command R+

Apr 24·104B

text

Open weight

Larger Command model for complex enterprise RAG and multi-step tool use.

Open details →

DeepSeek-V2

May 24·236B (21B active)

text

Open weight

Efficient MoE with Multi-head Latent Attention; strong open Chinese/English model.

→ 1 successor

Open details →

DeepSeek-V3

Dec 24·671B (37B active)

text

Open weight

Frontier open-weight MoE competitive with leading closed models on many benchmarks.

→ 2 successors

Open details →

Gemini 1.5

Feb 24·Undisclosed

multimodal

Commercial API

Long-context MoE Gemini with up to 1M–2M token windows.

→ 1 successor

Open details →

Gemini 2.0

Dec 24·Undisclosed

multimodal

Commercial API

Agent-oriented Gemini generation including Flash and experimental models.

→ 1 successor

Open details →

Llama 3

Apr 24·8B–70B

text

Open weight

Next-gen open Llama with strong post-training; foundation for 3.1–3.3 and ecosystem fine-tunes.

→ 1 successor

Open details →

Llama 3.1

Jul 24·8B–405B

text

Open weight

Long-context Llama including 405B open-weight flagship.

→ 3 successors

Open details →

Mistral Large

Feb 24·Undisclosed

text

Commercial API

Flagship commercial Mistral model for enterprise reasoning and multilingual use.

→ 1 successor

Open details →

Mistral Large 2

Jul 24·123B

text

Open weight

123B open-weight capable model with 128k context.

→ 2 successors

Open details →

GPT-4o

May 24·Undisclosed

multimodal

Commercial API

Omni model with native audio/vision/text and lower latency; default ChatGPT workhorse.

→ 2 successors

Open details →

o1

Dec 24·Undisclosed

text

Commercial API

OpenAI reasoning model with extended chain-of-thought compute at inference time.

Open details →

Stable Diffusion 3

Apr 24·800M–8B

vision

Open weight

Next-gen SD with Multimodal Diffusion Transformer architecture and improved text rendering.

Open details →

GLM-4

Jan 24·Undisclosed

text

Commercial API

Zhipu flagship bilingual model series competitive in Chinese enterprise AI.

→ 1 successor

Open details →

Qwen3

Apr 25·0.6B–235B-A22B

text

Open weight

Hybrid thinking Qwen generation spanning dense and MoE open models.

→ 1 successor

Open details →

Claude 4

May 25·Undisclosed

multimodal

Commercial API

Claude 4 Opus/Sonnet generation focused on agentic coding and extended thinking.

→ 2 successors

Open details →

DeepSeek-R1

Jan 25·671B (37B active)

text

Open weight

Open reasoning model that popularized RL-trained chain-of-thought at scale.

Open details →

Gemini 2.5

Mar 25·Undisclosed

multimodal

Commercial API

Thinking-model Gemini generation with strong coding and reasoning performance.

→ 1 successor

Open details →

Llama 4

Apr 25·Scout / Maverick MoE

multimodal

Open weight

Llama 4 MoE generation (Scout & Maverick) with natively multimodal training.

Open details →

GPT-4.1

Apr 25·Undisclosed

multimodal

Commercial API

GPT-4 series update focused on coding, instruction following, and very long context.

→ 1 successor

Open details →

Claude Opus 4.6

Feb 26·Undisclosed

multimodal

Commercial API

Anthropic Opus 4.6 — highest-capability Claude tier for complex coding and research workflows.

→ 1 successor

Open details →

Claude Sonnet 4.6

Feb 26·Undisclosed

multimodal

Commercial API

Balanced Claude Sonnet 4.6 tier for everyday coding, chat, and computer-use agents.

→ 1 successor

Open details →

Claude Fable 5

Jun 26·Undisclosed

multimodal

Commercial API

Anthropic Fable 5 — Mythos-class generally available Claude with 1M context; briefly export-controlled then restored globally in early July 2026.

Open details →

Claude Opus 4.8

Jun 26·Undisclosed

multimodal

Commercial API

Opus 4.8 upgrade with longer context and stronger multi-file software engineering performance.

→ 2 successors

Open details →

Claude Sonnet 5

Jun 26·Undisclosed

multimodal

Commercial API

Sonnet 5 — near-Opus balanced Claude default for free/Pro products and developer workloads.

Open details →

Claude Opus 5

Jul 26·Undisclosed

multimodal

Commercial API

Claude Opus 5 — new mid-flagship with effort dial (low→max), near-Fable-5 agentic coding at Opus pricing.

Open details →

GPT-5.3

Jan 26·Undisclosed

multimodal

Commercial API

OpenAI GPT-5.3 generation — stronger agentic coding and long-horizon tool use over the 4.1 line.

→ 1 successor

Open details →

GPT-5.4

Mar 26·Undisclosed

multimodal

Commercial API

GPT-5.4 mid-cycle upgrade focused on reliability, lower latency, and improved vision-grounded reasoning.

→ 1 successor

Open details →

GPT-5.5

May 26·Undisclosed

multimodal

Commercial API

GPT-5.5 with stronger realtime multimodal stack and enterprise control surfaces.

→ 1 successor

Open details →

GPT-5.6

Jul 26·Undisclosed

multimodal

Commercial API

GPT-5.6 generation shipped as durable Sol / Terra / Luna tiers — flagship reasoning, balanced mid-tier, and fast low-cost — for agents and production apps.

Open details →

Qwen3.5

Feb 26·Multi-size family

text

Open weight

Qwen3.5 open family — improved thinking modes and agent tool-use over Qwen3.

→ 1 successor

Open details →

Qwen3.7

Jun 26·Multi-size family

multimodal

Open weight

Qwen3.7 multimodal open release with stronger vision-language and coding stacks.

→ 1 successor

Open details →

Qwen3.8

Jul 26·2.4T MoE + 27B dense

multimodal

Open weight

Qwen3.8 generation — API-only Max preview in July; 2.4T open MoE weights and dense 27B multimodal land in mid-August.

Open details →

DeepSeek-V4

Apr 26·~1.6T MoE

text

Open weight

DeepSeek-V4 open-weight MoE preview generation succeeding V3 — later superseded by V4-Pro GA builds.

→ 2 successors

Open details →

DeepSeek-V4-Pro

Jul 26·1.6T MoE

text

Open weight

DeepSeek-V4-Pro flagship — MIT open weights, agent upgrades, flexible reasoning effort; 0813 GA checkpoint in August.

Open details →

DeepSeek-V4-Flash

Jul 26·Undisclosed MoE

text

Open weight

DeepSeek-V4-Flash — fast, cheap open-weight workhorse tier superseding earlier Flash previews.

Open details →

Gemini 3.1

Mar 26·Undisclosed

multimodal

Commercial API

Gemini 3.1 — Google DeepMind flagship with deeper thinking modes and native long video/context.

→ 1 successor

Open details →

Gemini 3.6 Flash

Jul 26·Undisclosed

multimodal

Commercial API

Google's July 2026 workhorse Flash — shipped with Flash-Lite and gated Flash Cyber while Gemini 3.5 Pro remained in partner testing.

Open details →

Muse Spark 1.1

Jul 26·Undisclosed

multimodal

Commercial API

Meta Muse Spark 1.1 — first major paid Meta agentic multimodal update with aggressive developer API pricing.

→ 1 successor

Open details →

Muse Spark 1.2

Aug 26·Undisclosed

multimodal

Commercial API

Meta Muse Spark 1.2 — August workhorse multimodal update; Meta signaled open weights for the Spark line.

Open details →

MiniMax M2.7

Feb 26·Undisclosed

text

Commercial API

MiniMax M2.7 text foundation model for chat, coding, and product integrations.

→ 1 successor

Open details →

MiniMax M3

Jun 26·Undisclosed

multimodal

Commercial API

MiniMax M3 multimodal generation for agents, video, and high-throughput apps.

Open details →

Mistral Small 4

Mar 26·Small class

text

Open weight

Efficient Mistral Small 4 open-weight model for edge and high-throughput European deployments.

Open details →

Mistral Medium 3.5

May 26·Medium class

multimodal

Open weight

Mistral Medium 3.5 multimodal mid-tier for enterprise RAG and European sovereign stacks.

Open details →

Kimi K2.6

Jan 26·Undisclosed

text

Commercial API

Moonshot Kimi K2.6 — ultra-long-context assistant for research and codebases.

→ 1 successor

Open details →

Kimi K3

Jul 26·2.8T MoE

multimodal

Open weight

Moonshot Kimi K3 — 2.8T open-weight MoE with native text/image/video and million-token context; led frontend coding arenas on debut.

Open details →

Grok 4.5

Jul 26·~1.5T

multimodal

Commercial API

xAI Grok 4.5 — frontier coding/agent flagship co-trained with Cursor session data; 500K context at aggressive $2/$6 pricing.

→ 1 successor

Open details →

Grok 4.6

Aug 26·~1.5T

multimodal

Commercial API

xAI Grok 4.6 — post-training upgrade for long-running agents, coding, and visual/design work; matches GPT-5.6 Sol-class composite scores at roughly one-third the price.

Open details →

EXAONE 4.5

Mar 26·Multi-size family

text

Open weight

LG AI Research EXAONE 4.5 open-weight bilingual models for Korean/English applications.

Open details →

LFM 2.5

Apr 26·Multi-size LFM

text

Open weight

Liquid AI LFM 2.5 — efficient foundation models optimized for edge and low-latency deployment.

Open details →

OpenEuroLLM

May 26·Multi-size family

text

Open weight

OpenEuroLLM consortium open multilingual models for European languages and sovereign AI stacks.

Open details →

Sarvam 2026

Apr 26·Indic multi-size

text

Open weight

Sarvam 2026 Indic foundation models for Indian languages, speech-adjacent text, and public-sector AI.

Open details →

Step 3.5

Apr 26·Undisclosed

multimodal

Commercial API

StepFun Step 3.5 multimodal model emphasizing perception and interactive agents.

Open details →

Inkling

Jul 26·975B MoE

text

Open weight

Thinking Machines Lab's first open-weight MoE — Mira Murati's lab debuts with Apache-2.0 Inkling for coding and agents.

Open details →

MiMo

May 26·Multi-size family

multimodal

Open weight

Xiaomi MiMo line targeting on-device intelligence and phone-native multimodal assistants.

Open details →

GLM-5

Mar 26·Undisclosed

multimodal

Commercial API

Zhipu GLM-5 bilingual flagship for enterprise agents and Chinese/English applications.

Open details →