Home β€Ί Artificial Intelligence (AI) / Machine Learning (ML) β€Ί Timeline of AI Models Cheat Sheet
Artificial Intelligence (AI) / Machine Learning (ML) Β· Sep 14, 2026

Timeline of AI Models Cheat Sheet

Below is a timeline of all AI models from some of the major vendors. Note, data may be incomplete or inaccurate. Assume up-to the publishing date.

Date Vendor Model / family Type Strengths / expertise
Jun-18 🟒 OpenAI GPT-1 🧠 General Β· πŸ”¬ Research The foundational Generative Pre-trained Transformer: demonstrated transfer learning from large-scale language pretraining.
Feb-19 🟒 OpenAI GPT-2 🧠 General Major leap in generative text quality Β· ✍️ Writing Β· πŸ’¬ Conversation. Initially released cautiously because of misuse concerns.
May-20 🟒 OpenAI GPT-3 🧠 General πŸš€ Major scaling milestone Β· ✍️ Writing Β· πŸ’» Code generation Β· few-shot learning.
Nov-20 🟒 OpenAI DALLΒ·E 🎨 Image Β· πŸ”¬ Research Early major text-to-image model demonstrating image generation from natural-language prompts.
Jan-21 🟒 OpenAI Codex πŸ’» Coding Natural-language-to-code generation; strategically important precursor to modern AI coding assistants.
Jan-21 🟒 OpenAI DALL·E public research announcement 🎨 Image Text-to-image generation became a major strategic model category.
Jun-21 🟒 OpenAI Codex API / Codex family πŸ’» Coding πŸ’» Software generation Β· natural-language programming Β· foundation for GitHub Copilot-era workflows.
Apr-22 🟒 OpenAI DALL·E 2 🎨 Image Major improvement in image realism, editing and variation generation.
Nov-22 🟒 OpenAI Whisper πŸŽ™οΈ Voice Β· πŸ”“ Open-weight Strategically important speech-recognition family Β· transcription Β· multilingual speech recognition.
Nov-22 🟒 OpenAI GPT-3.5 / ChatGPT generation 🧠 General Β· πŸ’¬ Conversation πŸ’¬ Conversational AI breakthrough that brought LLMs into mainstream consumer use.
Nov-22 🟦 Microsoft VALL-E πŸŽ™οΈ Voice Β· πŸ”¬ Research Important neural text-to-speech and voice-generation research model.
Mar-23 🟒 OpenAI GPT-4 🌐 Multimodal Β· 🧠 General Major frontier-model leap Β· 🧠 reasoning Β· πŸ’» coding Β· πŸ‘οΈ vision capability in multimodal versions.
Mar-23 🟦 Microsoft Kosmos-1 🌐 Multimodal Β· πŸ”¬ Research Microsoft’s early multimodal language-model research combining language and perception.
Mar-23 🟠 Anthropic Claude 1 🧠 General Β· πŸ’¬ Conversation Anthropic’s first major public Claude generation; emphasis on helpfulness and safety.
Apr-23 🟒 OpenAI GPT-3.5 Turbo 🧠 General · ⚑ Efficient Faster, lower-cost production/API-oriented GPT-3.5 generation.
Apr-23 🟧 Amazon / AWS Amazon Titan foundation models πŸ’Ό Professional/enterprise AWS’s strategically important in-house foundation-model family for Bedrock and enterprise applications.
Jul-23 🟠 Anthropic Claude 2 🧠 General Β· πŸ“š Long-context Major context-window expansion Β· writing Β· analysis Β· enterprise use.
Jul-23 πŸ”· Meta Llama 2 πŸ”“ Open-weight Β· 🧠 General Major open-model ecosystem milestone Β· 🧩 customization Β· 🏠 self-hosting.
Sep-23 🟒 OpenAI DALL·E 3 🎨 Image Much stronger prompt following and integration with conversational generation.
Sep-23 🟠 Anthropic Claude 2.1 🧠 General Β· πŸ“š Long-context Improved reliability and a strategically important 200K-token context window.
Sep-23 🟦 Microsoft Phi-1 / Phi-1.5 family ⚑ Small/efficient Demonstrated Microsoft’s strategy around surprisingly capable small language models.
Nov-23 🟧 Amazon / AWS Titan Image Generator 🎨 Image AWS-native image-generation foundation model for Bedrock.
Nov-23 🟧 Amazon / AWS Titan Multimodal Embeddings 🌐 Multimodal Β· πŸ’Ό Enterprise Multimodal retrieval and enterprise search/RAG applications. (Amazon Web Services, Inc.)
Dec-23 πŸ”΅ Google / Google DeepMind Gemini 1.0 🌐 Multimodal Β· 🧠 General Google’s new flagship multimodal family, including Ultra, Pro and Nano tiers.
Dec-23 πŸ”΅ Google / Google DeepMind Gemini Nano ⚑ Small/efficient Β· πŸ“± On-device Strategically important efficient/on-device AI direction.
Dec-23 🟦 Microsoft Phi-2 ⚑ Small/efficient Strong small-model capability for its size; important for local and constrained deployment.
Feb-24 πŸ”΅ Google / Google DeepMind Gemma πŸ”“ Open-weight Google’s major open-model family, optimized for customization and developer deployment.
Feb-24 🟒 OpenAI Sora πŸŽ₯ Video Landmark text-to-video model announcement; demonstrated long-form generative video capabilities.
Feb-24 🟒 OpenAI GPT-4 Turbo 🧠 General Β· πŸ“š Long-context Faster/cheaper GPT-4 generation with expanded context and stronger production usability.
Mar-24 🟠 Anthropic Claude 3 Haiku ⚑ Small/efficient Speed and affordability for high-volume workloads.
Mar-24 🟠 Anthropic Claude 3 Sonnet 🧠 General Β· πŸ’Ό Professional Balanced capability, speed and cost for professional work.
Mar-24 🟠 Anthropic Claude 3 Opus πŸ† Frontier/high-end Anthropic’s highest-capability Claude 3 model Β· reasoning Β· writing Β· analysis.
Apr-24 πŸ”· Meta Llama 3 πŸ”“ Open-weight Β· 🧠 General Major open-weight release that significantly strengthened the Llama ecosystem.
Apr-24 🟦 Microsoft Phi-3 family ⚑ Small/efficient Β· πŸ”“ Open-weight Small models for edge, local and enterprise deployment; important strategic expansion of Phi.
May-24 🟧 Amazon / AWS Titan Text Premier πŸ’Ό Professional/enterprise Β· πŸ“š RAG Amazon’s most advanced Titan text model, optimized for enterprise RAG and agent applications. (Amazon Web Services, Inc.)
May-24 🟒 OpenAI GPT-4o 🌐 Multimodal Β· πŸŽ™οΈ Voice Major “omni” model milestone: text, vision and near-real-time audio interaction in one family.
Jun-24 🟠 Anthropic Claude 3.5 Sonnet 🧠 General Β· πŸ’» Coding Major jump in πŸ’» coding, vision and professional reasoning; became highly influential for agentic coding.
Jun-24 πŸ”΅ Google / Google DeepMind Gemini 1.5 Pro 🌐 Multimodal Β· πŸ“š Long-context Major long-context milestone, including the 1M-token context direction.
Jun-24 πŸ”΅ Google / Google DeepMind Gemini 1.5 Flash ⚑ Small/efficient Β· 🌐 Multimodal Faster, cheaper multimodal model for scaled applications.
Jul-24 πŸ”· Meta Llama 3.1 πŸ”“ Open-weight Β· 🧠 General Expanded open-weight frontier with major larger-scale models and enterprise/self-hosting momentum.
Jul-24 🟒 OpenAI GPT-4o mini ⚑ Small/efficient · 🌐 Multimodal Cost-efficient model designed to bring capable multimodal AI to high-volume applications.
Aug-24 🟒 OpenAI GPT-4.5 preview-era research trajectory 🧠 General Not counted here as a 2024 public GPT-4.5 release; the verified GPT-4.5 public release belongs later.
Sep-24 🟒 OpenAI o1-preview 🧠 Reasoning OpenAI’s major public reasoning-model transition: extended inference for harder math, science and coding problems.
Sep-24 🟒 OpenAI o1-mini 🧠 Reasoning Β· πŸ’» Coding More efficient reasoning model, especially important for STEM and coding workloads.
Sep-24 πŸ”· Meta Llama 3.2 πŸ”“ Open-weight Β· 🌐 Multimodal Β· πŸ“± Edge Added strategically important vision models and smaller edge/mobile-oriented models.
Oct-24 🟠 Anthropic Claude 3.5 Haiku ⚑ Small/efficient Β· πŸ’» Coding Faster, lower-cost Claude optimized for production-scale workloads.
Oct-24 🟠 Anthropic Claude 3.5 Sonnet (new) πŸ’» Coding Β· πŸ€– Agentic Important upgrade in coding, computer interaction and agent-style workflows.
Oct-24 🟒 OpenAI GPT-Realtime / gpt-realtime πŸŽ™οΈ Voice Β· 🌐 Multimodal Native real-time speech interaction; strategically important for low-latency voice agents.
Dec-24 🟒 OpenAI o1 🧠 Reasoning Production reasoning model succeeding the preview generation; 🧠 reasoning Β· βž— mathematics Β· πŸ’» coding.
Dec-24 🟧 Amazon / AWS Amazon Nova family 🌐 Multimodal Β· πŸ’Ό Professional/enterprise Nova Micro, Lite, Pro and Canvas/Reel families established AWS’s new broad foundation-model strategy. (Amazon Web Services, Inc.)
Dec-24 πŸ”΄ DeepSeek DeepSeek-V3 πŸ”“ Open-weight Β· 🧠 General Β· πŸ’° Cost efficiency Large MoE open model emphasizing strong capability, speed and cost efficiency. (DeepSeek)
Jan-25 πŸ”΄ DeepSeek DeepSeek-R1 🧠 Reasoning Β· πŸ”“ Open-weight Major global reasoning/open-model milestone Β· 🧠 reasoning Β· βž— mathematics Β· πŸ’» coding. Released with MIT licensing and distilled variants. (DeepSeek)
Feb-25 🟒 OpenAI o3-mini 🧠 Reasoning Β· πŸ’» Coding More accessible and efficient advanced reasoning.
Feb-25 🟠 Anthropic Claude 3.7 Sonnet 🧠 Reasoning Β· πŸ’» Coding Β· πŸ€– Agentic Hybrid reasoning approach with extended thinking; major step toward modern coding agents.
Feb-25 🟒 OpenAI GPT-4.5 🧠 General · ✍️ Writing Large-scale general-purpose GPT release emphasizing natural interaction and knowledge work.
Mar-25 πŸ”΅ Google / Google DeepMind Gemma 3 πŸ”“ Open-weight Β· 🌐 Multimodal Google’s next major open-model generation with stronger multimodal capabilities.
Mar-25 πŸ”΅ Google / Google DeepMind Gemini 2.5 Pro 🧠 Reasoning Β· 🌐 Multimodal Major reasoning-focused Gemini generation with “thinking” capabilities.
Apr-25 🟒 OpenAI o3 🧠 Reasoning Β· πŸ€– Agentic High-end reasoning model with strong tool use, coding and complex multi-step problem solving.
Apr-25 🟒 OpenAI o4-mini 🧠 Reasoning Β· πŸ’» Coding Efficient reasoning model with strong STEM and tool-use capability.
Apr-25 🟧 Amazon / AWS Amazon Nova Premier πŸ† Frontier/high-end Β· 🌐 Multimodal Β· πŸ“š Long-context AWS’s most capable Nova model for complex planning, tool use and million-token-scale context. (Amazon Web Services, Inc.)
Apr-25 πŸ”· Meta Llama 4 πŸ”“ Open-weight Β· 🌐 Multimodal Major new Llama generation emphasizing multimodality and MoE-scale architectures.
May-25 🟠 Anthropic Claude Sonnet 4 πŸ’» Coding Β· πŸ€– Agentic Strong software engineering and agentic task execution.
May-25 🟠 Anthropic Claude Opus 4 πŸ† Frontier/high-end Β· πŸ’» Coding Β· πŸ€– Agentic High-end agentic coding and long-running task execution.
Jun-25 🟒 OpenAI GPT-Image-1 / GPT Image family 🎨 Image OpenAI’s strategically important modern image-generation API family.
Jun-25 🟒 OpenAI Codex agent generation πŸ’» Coding Β· πŸ€– Agentic Modern coding-agent evolution: autonomous software tasks rather than simply code completion.
Aug-25 🟠 Anthropic Claude Opus 4.1 πŸ’» Coding Β· 🧠 Reasoning Incremental but strategically significant Opus improvement for advanced coding and agentic work.
Aug-25 πŸ”΄ DeepSeek DeepSeek-V3.1 πŸ€– Agentic Β· 🧠 Reasoning Β· πŸ”“ Open-weight Hybrid Think/Non-Think inference and stronger tool use; DeepSeek explicitly positioned it as a step toward the agent era. (DeepSeek)
Sep-25 🟠 Anthropic Claude Sonnet 4.5 πŸ’» Coding Β· πŸ€– Agentic Major Sonnet-class upgrade for autonomous coding and professional workflows.
Sep-25 πŸ”΄ DeepSeek DeepSeek-V3.2-Exp πŸ”“ Open-weight Β· πŸ“š Long-context Experimental DeepSeek Sparse Attention architecture targeting more efficient long-context inference.
Oct-25 🟠 Anthropic Claude Haiku 4.5 ⚑ Small/efficient Β· πŸ’» Coding Fast and economical modern Claude generation.
Nov-25 🟠 Anthropic Claude Opus 4.5 πŸ† Frontier/high-end Β· πŸ’» Coding Β· πŸ€– Agentic Strong frontier agentic and coding capability.
Nov-25 πŸ”΅ Google / Google DeepMind Gemini 3 / Gemini 3 Pro 🧠 Reasoning Β· 🌐 Multimodal Β· πŸ€– Agentic Major new Gemini generation combining advanced reasoning, multimodality and agentic capability. (blog.google)
Dec-25 πŸ”΅ Google / Google DeepMind Gemini 3 Flash ⚑ Small/efficient Β· 🌐 Multimodal Β· 🧠 Reasoning High-speed model combining strong reasoning with image, audio and video understanding. (blog.google)
Dec-25 πŸ”΄ DeepSeek DeepSeek-V3.2 🧠 Reasoning Β· πŸ€– Agentic Β· πŸ”“ Open-weight Reasoning-first successor with integrated thinking during tool use. (DeepSeek)
Dec-25 πŸ”΄ DeepSeek DeepSeek-V3.2-Speciale 🧠 Reasoning Β· βž— Mathematics Β· πŸ”¬ Specialized High-end reasoning variant focused on extremely difficult mathematics and competition-style reasoning. (DeepSeek)
Dec-25 🟧 Amazon / AWS Nova 2 Sonic πŸŽ™οΈ Voice Β· 🌐 Multimodal General-availability speech-to-speech foundation model for real-time conversational AI. (Amazon Web Services, Inc.)
5-Feb-26 🟠 Anthropic Claude Opus 4.6 πŸ† Frontier/high-end Β· πŸ’» Coding Β· πŸ€– Agentic Better planning, longer-running agentic tasks, codebase work and a 1M-token context beta. (Anthropic)
######## 🟠 Anthropic Claude Sonnet 4.6 πŸ’» Coding Β· πŸ€– Agentic Β· πŸ“š Long-context Major Sonnet upgrade across coding, computer use, planning and knowledge work. (Anthropic)
Apr-26 πŸ”΄ DeepSeek DeepSeek-V4 Preview πŸ”“ Open-weight Β· πŸ“š Long-context Officially released/open-sourced preview emphasizing cost-effective million-token context. (DeepSeek)
######## 🟠 Anthropic Claude Opus 4.8 πŸ† Frontier/high-end Β· πŸ’» Coding Β· πŸ€– Agentic Improved coding, agentic skills and practical knowledge work; added more flexible effort/speed controls. (Anthropic)
######## πŸ”΅ Google / Google DeepMind Gemini 3.5 Flash πŸ€– Agentic Β· πŸ’» Coding Β· ⚑ Efficient Major transition toward frontier intelligence with action, optimized for long-horizon agentic workflows. (blog.google)
######## πŸ”΅ Google / Google DeepMind Gemini Omni Flash 🌐 Multimodal Β· πŸŽ₯ Video Native multimodal generation from mixed inputs, including video-oriented creation and editing. (blog.google)
24-Jun-26 πŸ”΅ Google / Google DeepMind Gemini 3.5 Flash Computer Use πŸ–₯️ Computer use Β· πŸ€– Agentic Computer interaction became integrated into the main Flash model for browser, mobile and desktop agents. (blog.google)
30-Jun-26 🟠 Anthropic Claude Sonnet 5 πŸ€– Agentic Β· πŸ’» Coding Β· 🧠 Reasoning Explicitly designed as Anthropic’s most agentic Sonnet model, with planning, browser and terminal tool use. (Anthropic)
9-Jul-26 🟒 OpenAI GPT-5.6 family 🧠 General Β· πŸ’Ό Professional Β· πŸ€– Agentic Sol, Terra and Luna tiers: stronger efficiency, coding, science, cybersecurity and multi-agent work. (OpenAI)
21-Jul-26 πŸ”΅ Google / Google DeepMind Gemini 3.6 Flash ⚑ Efficient Β· πŸ’» Coding Β· πŸ€– Agentic New Flash workhorse focused on token efficiency, lower latency and production agents. (blog.google)
21-Jul-26 πŸ”΅ Google / Google DeepMind Gemini 3.5 Flash-Lite ⚑ Small/efficient Lower-cost/high-throughput Gemini generation for scaled workloads. (blog.google)
24-Jul-26 🟠 Anthropic Claude Opus 5 πŸ† Frontier/high-end Β· πŸ’» Coding Β· πŸ’Ό Professional New flagship everyday Opus-class model with state-of-the-art coding and knowledge-work positioning. (Anthropic)
3-Sep-26 🟒 OpenAI GPT-6 Astra 🧠 Reasoning Β· πŸ–₯️ Computer use Β· πŸ€– Agentic Β· πŸ† Frontier OpenAI’s newest major verified model as of the cutoff: advanced computer use, browsing, software engineering, cybersecurity, science and professional work. (OpenAI)
######## πŸ”΄ DeepSeek DeepSeek-V4.1-Flash ⚑ Efficient Β· πŸ‘οΈ Vision Β· πŸ€– Agentic Β· πŸ”“ Open-weight New architecture family with native visual understanding, asymmetric MoE design and efficiency aimed at high-throughput agentic workloads. (DeepSeek)

Specialized AI Model Families

Vendor Family Primary specialty What it’s good for
🟒 OpenAI Whisper πŸŽ™οΈ Speech recognition Multilingual transcription and speech translation
🟒 OpenAI DALL·E 🎨 Image generation Text-to-image creation and editing
🟒 OpenAI GPT-image 🎨 Image Native multimodal image creation/editing and text rendering O OpenAI
🟒 OpenAI Sora πŸŽ₯ Video Generative video and video transformation
🟒 OpenAI Codex πŸ’» Coding agents Repository-scale development and autonomous software engineering
🟒 OpenAI GPT-Live πŸŽ™οΈ Voice Full-duplex real-time conversational AI O OpenAI
🟒 OpenAI GPT-6 Astra πŸ–₯️ Computer-use agents Computer operation, browsing, professional workflows, coding and research O OpenAI
🟠 Anthropic Claude Computer Use πŸ–₯️ Computer use Browser/desktop interaction and UI automation
🟠 Anthropic Claude Code πŸ’» Coding agents Long-running software engineering
🟠 Anthropic Claude Agent SDK πŸ€– Agent framework/model tooling Building autonomous agents with memory, permissions and tool use
πŸ”΅ Google Gemma πŸ”“ Open-weight / small Local inference, customization and edge deployment
πŸ”΅ Google Veo πŸŽ₯ Video Generative video
πŸ”΅ Google Gemini Image 🎨 Image Image generation/editing
πŸ”΅ Google Gemini Audio / Live πŸŽ™οΈ Audio Real-time voice and audio reasoning
πŸ”΅ Google Gemini Computer Use πŸ–₯️ Computer use Operating browser/software interfaces
πŸ”΅ Google Gemini 3.5 agentic family πŸ€– Agents Long-horizon action and tool execution B blog.google
πŸ”· Meta Llama Vision πŸ‘οΈ Vision Image understanding in customizable/open deployments M Meta AI
πŸ”· Meta Code Llama πŸ’» Coding Code generation and completion
πŸ”· Meta Llama Guard πŸ›‘οΈ Safety Content moderation and safety classification
πŸ”΄ DeepSeek DeepSeek-R1 🧠 Reasoning Math, science, coding and open-weight reasoning
πŸ”΄ DeepSeek DeepSeek-V3 family 🧠 General / πŸ’» Coding Efficient general intelligence and coding
πŸ”΄ DeepSeek DeepSeek-V4 🌐 Multimodal / πŸ€– Agents Long-context reasoning, vision and agentic work D DeepSeek API Docs
🟦 Microsoft Phi ⚑ Small models Efficient reasoning, edge and local deployment
🟦 Microsoft Phi reasoning 🧠 Small reasoning Math, science and coding on compact models
🟧 Amazon Titan πŸ’Ό Enterprise foundation models Text, embeddings, image generation and Bedrock applications
🟧 Amazon Nova 🌐 Multimodal Enterprise text, image/video and agent workloads A Amazon Web Services, Inc.
🟧 Amazon Nova 2 🧠 Reasoning / πŸ€– Agents Extended thinking, tools and 1M-token enterprise workflows A Amazon Web Services, Inc.
🟧 Amazon Nova Sonic πŸŽ™οΈ Voice Real-time speech-to-speech applications A Amazon Web Services, Inc.

What Each Vendor is Becoming Known By

Vendor πŸ† Core identity Particularly strong at
🟒 OpenAI Frontier general intelligence increasingly designed as an agentic computer operator 🧠 Reasoning Β· πŸ’» Coding Β· πŸ€– Agents Β· πŸ–₯️ Computer use Β· πŸŽ™οΈ Voice Β· 🎨 Image
🟠 Anthropic Coding and long-running professional agents πŸ’» Coding Β· πŸ€– Agents Β· πŸ–₯️ Computer use Β· πŸ“š Long context Β· πŸ’Ό Knowledge work
πŸ”΅ Google DeepMind Native multimodality + agents + full-stack AI 🌐 Multimodal Β· πŸ€– Agents Β· πŸ’» Coding Β· πŸŽ™οΈ Audio Β· 🎨 Image Β· πŸŽ₯ Video
πŸ”· Meta Open-weight/customizable AI ecosystem πŸ”“ Open weights Β· 🧩 Customization Β· πŸ“± Edge Β· 🌐 Multimodality
πŸ”΄ DeepSeek Efficient open-weight reasoning 🧠 Reasoning Β· βž— Math Β· πŸ’» Coding Β· πŸ’° Efficiency Β· πŸ”“ Open weights
🟦 Microsoft Small efficient models for local/enterprise computing ⚑ Small models Β· πŸ“± Edge Β· 🧠 Efficient reasoning Β· πŸ’Ό Enterprise
🟧 Amazon/AWS Cloud-delivered foundation models and enterprise agents ☁️ Bedrock Β· πŸ’Ό Enterprise Β· πŸ’° Price/performance Β· πŸŽ™οΈ Voice Β· πŸ€– Agents

 

Leave a Reply

Your email address will not be published. Required fields are marked *

Carl de Souza
KEEP LEARNING. KEEP BUILDING.

Explore AI, agents & Microsoft technology.

I share practical ideas, tutorials, and videos about AI, AI agents, Microsoft technologies, and the Power Platform.