A timeline of the biggest events in modern AI history — a running list of what mattered.

Illustration of major AI lab logos: OpenAI, Anthropic, DeepSeek, Google DeepMind, and xAI.
  1. Models

    OpenAI launches GPT-5.5

    OpenAI's latest frontier model—SOTA on most major benchmarks, with standout strength in agentic coding, computer use, and knowledge work.

    OpenAI opens in a new tab

  2. OpenAI launches GPT Image 2 (Images 2.0)

    GPT Image 2 is OpenAI’s flagship image model behind ChatGPT Images 2.0—stronger world knowledge and instruction following, dense text and layouts, optional thinking mode with reasoning and tools (including web-informed generation), and consistent multi-image batches; API snapshot dated Apr 21, 2026.

    OpenAI opens in a new tab

  3. Moonshot AI launches Kimi K2.6

    Moonshot AI published Kimi K2.6 as an open-source coding-focused model for long-horizon coding, coding-driven design, agent swarms, proactive agents, and Claw Groups.

    Moonshot AI opens in a new tab

  4. Anthropic launches Claude Opus 4.7

    Anthropic made Opus 4.7 generally available with stronger advanced software engineering, better high-resolution vision, and improved long-running task performance.

    Anthropic opens in a new tab
  5. 2

    Anthropic launches Claude Mythos Preview and announces Project Glasswing

    A SOTA model for advanced autonomous cybersecurity reasoning, released restrictively for safety, alongside Project Glasswing to secure open-source software with major industry partners.

    Anthropic opens in a new tab

    Z.ai launches GLM-5.1

    Z.ai released GLM-5.1 as its next-generation flagship for agentic engineering, with stronger coding, long-horizon tool use, and open weights under the MIT License.

    Z.ai opens in a new tab
    Models

    Anthropic launches Claude Mythos Preview and announces Project Glasswing

    A SOTA model for advanced autonomous cybersecurity reasoning, released restrictively for safety, alongside Project Glasswing to secure open-source software with major industry partners.

    Anthropic opens in a new tab

  6. OpenAI launches GPT-5.4

    Frontier model for professional work, released in ChatGPT, the API, and Codex with stronger reasoning, coding, computer use, and tool-use capabilities.

    OpenAI opens in a new tab

  7. Google DeepMind launches Gemini 3.1 Pro

    Google's latest flagship. More than doubles reasoning performance over Gemini 3 Pro. Released in preview via Gemini API, AI Studio, and Vertex AI.

  8. xAI launches Grok 4.2

    Beta release with rapid learning architecture — improves weekly via user feedback. 256K context, 4-agent parallel reasoning. Medical document analysis added.

    Anthropic launches Claude Sonnet 4.6

    Most capable Sonnet yet. Full upgrade across coding, computer use, long-context reasoning, agent planning, and design. 1M token context window in beta.

  9. Z.ai launches GLM-5

    China's first public AI company frontier model. Targets complex systems engineering and long-horizon agentic tasks.

  10. Anthropic launches Claude Opus 4.6

    Anthropic upgraded its smartest Opus model with stronger coding, longer agentic work, better code review and debugging, and a 1M token context window in beta.

    Anthropic opens in a new tab

    OpenAI launches GPT-5.3-Codex

    Most capable agentic coding model. Next-generation Codex line for agentic software engineering. Available via Codex app, CLI, and IDE extensions.

  11. OpenAI launches GPT-5.2 Codex

    OpenAI introduced GPT-5.2 Codex as its coding-focused model for agentic software engineering, with stronger long-running coding, review, and debugging workflows.

    OpenAI opens in a new tab

  12. Google DeepMind launches Gemini 3 Flash

    Fast frontier-class model rivaling larger models at a fraction of the cost. Default model in the Gemini app.

  13. Models

    OpenAI launches GPT-5.2, its next-generation frontier model

    A frontier model family for professional work and long-running agents, including Pro, Thinking, and base variants.

    OpenAI opens in a new tab

  14. Mistral AI launches Mistral Large 3

    Sparse MoE with 41B active params. Open weights. Strong reasoning and multilingual.

  15. Z.ai launches GLM-4.7

    Open-weights model from Z.ai topping global coding and reasoning leaderboards. Includes GLM-4.7 Flash variant.

    DeepSeek launches DeepSeek V3.2

    Further iteration on V3 series. Enhanced capabilities across all benchmarks.

  16. Models

    Anthropic launches Claude Opus 4.5, its most advanced model

    Anthropic's most capable model brought a reported 20% accuracy gain, Infinite Chats, and stronger Excel and financial-modeling workflows.

    Anthropic opens in a new tab
  17. Models

    Google DeepMind launches Gemini 3 Pro, its next-generation flagship model

    Google's most powerful model replaced the 2.5 series and claimed a 50%+ improvement over Gemini 2.5 Pro.

    Google DeepMind opens in a new tab

  18. xAI launches Grok 4.1

    Incremental update to Grok 4. Improved reasoning and instruction following. Available on X and via API.

  19. OpenAI launches GPT-5.1

    Family of four models with adaptive reasoning. Faster, more conversational, improved coding. Rolled out to all ChatGPT users.

  20. Moonshot AI launches Kimi K2.5

    Upgraded Kimi model with thinking and reasoning capabilities.

  21. Z.ai launches GLM-4.6

    First GLM model with native support for China's domestic chips (Cambricon, Moore Threads). FP8 and Int4 quantization.

  22. DeepSeek launches DeepSeek V3.1

    Major upgrade to V3. Improved reasoning and coding. Open weights.

  23. Z.ai launches GLM-4.5V

    Vision-language model from Z.ai. 106B parameters, strong multimodal understanding.

  24. Models

    OpenAI launches GPT-5, its next-generation flagship model

    A next-generation flagship with a major intelligence leap across coding, math, writing, and reasoning.

    OpenAI opens in a new tab
  25. 2

    OpenAI launches GPT-OSS, its open-weight model series

    OpenAI's first open-weight models since GPT-2 arrived in 20B and 120B variants, marking a historic open-source shift.

    OpenAI opens in a new tab

    Anthropic launches Claude Opus 4.1

    Upgrade to Claude 4. Improved coding, instruction following, and tool use.

    Models

    OpenAI launches GPT-OSS, its open-weight model series

    OpenAI's first open-weight models since GPT-2 arrived in 20B and 120B variants, marking a historic open-source shift.

    OpenAI opens in a new tab

  26. Anthropic launches Claude 4.5 Sonnet

    Newest Anthropic model. Improved creative writing, nuance, multi-step reasoning.

  27. Z.ai launches GLM-4.5

    Z.ai flagship open MoE. 355B total, strong reasoning, coding, and agentic capabilities. Claimed cheaper to run than DeepSeek.

  28. Alibaba launches Qwen3 Coder

    Dedicated coding model from Qwen. 480B MoE architecture with 35B active parameters.

  29. 2
    Models

    xAI launches Grok-4, its most powerful model

    xAI's most powerful model brought a major reasoning leap and was trained on an expanded Colossus cluster.

    xAI opens in a new tab
    Models

    Moonshot AI launches Kimi K2, its open-weight MoE model

    A massive one-trillion-parameter MoE with open weights, competitive with frontier models and a major Chinese AI milestone.

    Moonshot AI opens in a new tab

  30. OpenAI launches o3 Pro

    Most powerful OpenAI reasoning model. Extended thinking for frontier problems.

  31. Google DeepMind launches Gemini 2.5 Pro

    Latest 2.5 Pro. Enhanced coding, reasoning, agentic capabilities.

  32. 2

    Anthropic launches Claude 4 Opus, its most powerful model

    Anthropic's strongest model generation pushed parallel tool use and long autonomous tasks.

    Anthropic opens in a new tab

    Anthropic launches Claude 4 Sonnet

    Fast successor to 3.7 Sonnet. Excellent speed-intelligence balance.

    Models

    Anthropic launches Claude 4 Opus, its most powerful model

    Anthropic's strongest model generation pushed parallel tool use and long autonomous tasks.

    Anthropic opens in a new tab

  33. Google DeepMind launches Gemini 2.5 Flash

    Cost-efficient reasoning. Controllable thinking depth. #1 Chatbot Arena for speed.

  34. Product

    OpenAI launches Codex research preview, their agentic coding tool

    An agentic coding tool that can write, test, and fix code end-to-end, integrated into ChatGPT for Pro, Team, and Enterprise, laying the foundation for today's Codex.

    OpenAI opens in a new tab

  35. Mistral AI launches Mistral Medium 3

    Front-tier, competitive with GPT-4o. Apache 2.0. Strong multilingual.

  36. OpenAI launches o4-mini

    Efficient reasoning model. Best cost-performance for coding and STEM.

  37. OpenAI launches o3

    Full o3 reasoning model. Successor to o1. Deep chain-of-thought.

  38. OpenAI launches GPT-4.1

    Optimized for coding/instruction following. 1M context. 50% cheaper than GPT-4o.

  39. Models

    Meta launches LLaMA 4, its next-generation open-weight model

    Scout and Maverick brought native multimodality, with Scout reaching a 10-million-token context window.

    Meta AI opens in a new tab
  40. 2

    Google DeepMind launches Gemini 2.5 Pro, its next-generation LLM

    Gemini 2.5 Pro launched with built-in reasoning, a one-million-token context window, and the top LMArena position at launch.

    Google opens in a new tab

    OpenAI launches 4o image generation

    OpenAI shipped native image generation inside GPT-4o—useful, prompt-accurate, photorealistic output with strong text rendering, refinement across turns in chat, C2PA provenance, and rollout across ChatGPT tiers (developers gained API access shortly after).

    OpenAI opens in a new tab
    Models

    Google DeepMind launches Gemini 2.5 Pro, its next-generation LLM

    Gemini 2.5 Pro launched with built-in reasoning, a one-million-token context window, and the top LMArena position at launch.

    Google opens in a new tab

  41. Google DeepMind launches Gemma 3

    1B/4B/12B/27B. Multimodal (text+vision). Single GPU. 128K context.

  42. Mistral AI launches Mistral Small 3.1

    Adds vision capabilities to Small 3.0. Multimodal, 128K context. Apache 2.0.

  43. Alibaba launches QwQ-32B

    Dedicated reasoning model from Qwen team. Strong mathematical and logical reasoning. Apache 2.0.

  44. 2

    Anthropic launches Claude Code research preview, their agentic coding tool

    An agentic coding tool in the terminal that can read, modify, test, and execute across entire codebases, laying the foundation for today's Claude Code.

    Anthropic opens in a new tab

    OpenAI launches GPT-4.5

    Largest OpenAI model yet. Focus on EQ, creativity, reduced hallucinations.

    Product

    Anthropic launches Claude Code research preview, their agentic coding tool

    An agentic coding tool in the terminal that can read, modify, test, and execute across entire codebases, laying the foundation for today's Claude Code.

    Anthropic opens in a new tab

  45. Anthropic launches Claude 3.7 Sonnet

    Hybrid reasoning — toggle instant/extended thinking. Best coding model at launch.

  46. xAI launches Grok-3

    Trained on Colossus supercluster (100K GPUs). Strong reasoning capabilities.

  47. Models

    DeepSeek launches DeepSeek-R1, its open-weights reasoning model

    An open reasoning model rivaling o1, built around a pure RL approach and strong enough to send shockwaves through the market.

    DeepSeek opens in a new tab
  48. Models

    DeepSeek launches DeepSeek V3, its MoE-based LLM

    A 671B-parameter MoE trained for a reported $5.5M and competitive with GPT-4o and Claude 3.5 Sonnet, reshaping the cost-efficiency debate.

    DeepSeek opens in a new tab

  49. Google DeepMind launches Gemini 2.0 Flash Thinking

    Experimental reasoning model with visible chain-of-thought. Google's answer to o1.

  50. Google DeepMind launches Gemini 2.0 Flash

    Multimodal output (text, images, audio). Built for agentic AI. 2x faster than 1.5 Pro.

  51. Product

    OpenAI releases Sora, bringing its AI video from research preview to the public

    Sora 1 set a new standard for text-to-video with strong realism and motion, opening broad access and accelerating the race between OpenAI and rivals such as Google’s later Veo in generative video.

    OpenAI opens in a new tab

  52. OpenAI launches o1-pro

    Enhanced reasoning model with more compute for complex tasks. Available in ChatGPT Pro tier.

  53. Platform

    Anthropic launches Model Context Protocol, its open-source standard for tool use

    An open-source protocol for connecting LLMs to tools, data, and systems, standardizing how models access external context across the AI ecosystem.

    Anthropic opens in a new tab
  54. Models

    OpenAI launches o1-preview, its first reasoning model

    The first OpenAI reasoning model put inference-time reasoning in the spotlight and targeted PhD-level science and math tasks.

    OpenAI opens in a new tab
  55. Models

    xAI launches Grok-2, its next-generation model

    Competitive with GPT-4o and Claude 3.5 and available on the X platform.

    xAI opens in a new tab

  56. Mistral AI launches Mistral Large 2

    128K context. Competitive with GPT-4o and Llama 3.1 405B. 12 languages.

  57. Models

    Meta launches LLaMA 3.1, its upgraded open-weight model

    The 405B model became the largest open model of its moment, with 128K context and performance near GPT-4 on many benchmarks.

    Meta AI opens in a new tab

  58. Google DeepMind launches Gemma 2

    9B and 27B sizes. Outperforms models 2x its size. Knowledge distillation from Gemini.

  59. Models

    Anthropic launches Claude 3.5 Sonnet, its upgraded model

    Claude 3.5 Sonnet surpassed GPT-4o and Gemini 1.5 Pro on several benchmarks while running twice as fast as Claude 3 Opus at lower cost.

    Anthropic opens in a new tab

  60. DeepSeek launches DeepSeek Coder V2

    First open MoE code model matching GPT-4 Turbo on coding. 338 programming languages.

  61. Zhipu AI launches GLM-4

    Open-weight GLM-4 series (incl. 9B): 128K context, 26 languages — competitive with Llama 3 8B.

  62. Models

    OpenAI launches GPT-4o, its omni model

    An omni model with native audio, vision, and text that was twice as fast and 50% cheaper than GPT-4 Turbo.

    OpenAI opens in a new tab

  63. DeepSeek launches DeepSeek V2

    236B MoE with only 21B active. Multi-head Latent Attention for efficiency.

  64. Models

    Meta launches LLaMA 3, its next-generation open-weight model

    Trained on 15 trillion tokens and released in 8B and 70B sizes, LLaMA 3 became a new open-source state of the art with massive adoption.

    Meta AI opens in a new tab

  65. Mistral AI launches Mixtral 8x22B

    Large MoE with 8 experts of 22B parameters. Strong multilingual and code performance. Open weights.

  66. Models

    xAI launches Grok-1, its first open-source model

    xAI's first open-source model was a 314B-parameter MoE released under Apache 2.0 and among the largest open MoE models at the time.

    xAI opens in a new tab
  67. Models

    Anthropic launches Claude 3, its next-generation model family

    Haiku, Sonnet, and Opus brought 200K context and vision, with Opus matching GPT-4 on many benchmarks.

    Anthropic opens in a new tab

  68. Mistral AI launches Mistral Large

    Mistral's first flagship commercial model. 32K context. Top-tier reasoning.

  69. Google DeepMind launches Gemma

    Google's open-source model from Gemini research. 2B and 7B sizes. Strong for its class.

  70. Models

    Google DeepMind launches Gemini 1.5 Pro, its next-generation model

    A one-million-token context window, MoE architecture, and whole-codebase processing made it a major leap in long-context modeling.

    Google opens in a new tab

  71. Google launches Gemini 1.0 Ultra

    Most capable Gemini 1.0. Beat GPT-4 on 30/32 benchmarks. Powers Gemini Advanced.

  72. Mistral AI launches Mixtral 8x7B

    Open-source mixture of experts. Matched GPT-3.5 quality with only 12.9B active params. Game-changer.

  73. Models

    Google DeepMind launches Gemini 1.0, its multimodal model family

    Google's Gemini family arrived in Nano, Pro, and Ultra sizes and was designed as natively multimodal from training.

    Google opens in a new tab

  74. OpenAI launches DALL·E 3

    DALL·E 3 shipped in ChatGPT Plus and Enterprise: create and iterate on images in chat, with a model that follows long, detailed prompts more closely, improved rendering of text and fine detail, and responsible-deployment mitigations before broad use.

    OpenAI opens in a new tab

  75. Mistral AI launches Mistral 7B

    Mistral 7B outperformed Llama 2 70B on common benchmarks despite being much smaller, introduced grouped-query sliding-window attention, and shipped under Apache 2.0 — Mistral AI’s first open-weight release.

  76. Models

    Meta launches LLaMA 2, its open-weight model for commercial use

    One of the first major open-weight model releases for commercial use, with 7B, 13B, and 70B sizes plus RLHF-tuned chat variants.

    Meta AI opens in a new tab
  77. Models

    Anthropic launches Claude 2

    Claude 2 expanded Anthropic's public model lineup with stronger capabilities and broader access through chat and API surfaces.

    Anthropic opens in a new tab
  78. 2
    Models

    Anthropic launches Claude

    Claude and Claude Instant became Anthropic's first broadly introduced assistant models and a direct ChatGPT competitor.

    Anthropic opens in a new tab
    Models

    OpenAI launches GPT-4, its first multimodal model

    GPT-4 could accept both text and images, raising the bar for frontier-model reasoning and multimodal capabilities.

    OpenAI opens in a new tab
  80. Product

    Anysphere launches Cursor, their AI-native code editor

    A very early version of Cursor put AI-native chat, edits, and codebase reasoning inside the IDE; it later grew into one of the leading developer tools and helped kickstart vibe coding.

    Cursor opens in a new tab
  81. Models

    Meta launches LLaMA, its research LLM

    LLaMA 1 and its leaked weights ignited the open-source LLM wave and showed smaller models could approach GPT-3-class performance.

    Meta AI opens in a new tab
  82. Product

    OpenAI launches ChatGPT, its chat-based AI assistant

    GPT-3.5 with RLHF in a chat interface reached 100 million users in two months and helped define the modern AI era.

    OpenAI opens in a new tab

  83. OpenAI launches DALL·E 2

    OpenAI published the two-stage system behind DALL·E 2: a model generates CLIP image embeddings from a text caption, and a diffusion decoder turns them into images—sharper, more photoreal results and a big step up from the original DALL·E.

    OpenAI opens in a new tab
  84. Models

    DeepMind launches Chinchilla, its compute-optimal model

    Proved that smaller models trained on more data can outperform larger undertrained ones, redefining scaling laws.

    Google DeepMind opens in a new tab
  85. Models

    OpenAI launches InstructGPT, its alignment-trained model

    Introduced RLHF for alignment and pioneered the technique of training models to follow human instructions safely.

    OpenAI opens in a new tab
  86. Platform

    Anthropic is founded by former OpenAI leaders

    Dario and Daniela Amodei and a group of former OpenAI colleagues (reported as seven co-founders, including Jared Kaplan, Jack Clark, and Chris Olah) form Anthropic as a public benefit corporation, prioritizing safety after strategic tension with OpenAI’s direction on speed and commercialization.

    Anthropic opens in a new tab
  87. Product

    OpenAI introduces DALL·E, text-to-image generation

    A 12-billion-parameter transformer trained on text–image pairs to synthesize images from natural-language captions—combining disparate concepts, controlling attributes and viewpoint, and inpainting regions—showing that rich visual compositions can be steered with words.

    OpenAI opens in a new tab
  88. Models

    OpenAI releases GPT-2

    OpenAI releases GPT-2 but withholds the largest version at launch due to misuse concerns. The decoder-only transformer is trained with next-token prediction to generate coherent long-form text.

    OpenAI opens in a new tab
  89. Models

    OpenAI introduces GPT-1

    OpenAI reveals the first Generative Pre-trained Transformer (GPT), demonstrating unsupervised pre-training followed by supervised fine-tuning for strong downstream NLP performance.

    OpenAI opens in a new tab
  90. Research paper

    PPO

    OpenAI introduces Proximal Policy Optimization, a simpler and more stable policy gradient method that would become widely used across many reinforcement learning domains, including RLHF.

    arXiv opens in a new tab
  91. 2
    Research paper

    Attention is All You Need

    Google introduces the Transformer architecture, a breakthrough deep learning architecture based on the attention mechanism. The architecture shows strong gains on language translation tasks.

    arXiv opens in a new tab
    Research paper

    RLHF

    Christiano et al. publish the technique of reinforcement learning from human feedback (RLHF), which would later be used extensively to align large language models.

    arXiv opens in a new tab
  92. Platform

    PyTorch

    Facebook releases PyTorch, a Python-first deep learning framework that would eventually become the dominant framework for AI research.

    PyTorch opens in a new tab
  93. Research paper

    AlphaGo

    DeepMind's AlphaGo defeats top human player Lee Sedol in the board game Go, defying what many considered possible.

    Google DeepMind opens in a new tab
  94. Platform

    OpenAI is founded and launched as a nonprofit

    Announced as a nonprofit with co-chairs Sam Altman and Elon Musk, CTO Greg Brockman, research director Ilya Sutskever, and founding researchers including Andrej Karpathy, John Schulman, Durk Kingma, and Wojciech Zaremba—committed to digital intelligence that benefits humanity and broad publication of results.

    OpenAI opens in a new tab
  95. Platform

    TensorFlow

    Google open-sources TensorFlow, its internal deep learning framework. Initially developed by the Google Brain team, TensorFlow would become one of the most influential AI frameworks.

    Google Research opens in a new tab
  96. Platform

    DeepMind is founded

    Demis Hassabis, Shane Legg, and Mustafa Suleyman start DeepMind to pursue general AI via deep reinforcement learning and games; Google acquired the lab in 2014, and in April 2023 it merged with Google Brain as Google DeepMind.

    Google DeepMind opens in a new tab