History of Artificial Intelligence

Difference Engine – Babbage designs and presents a mechanical machine for computing mathematical tables (e.g. logarithms) without human error. Funding and technology of the time did not allow a full version to be completed; parts were built later in museums.

Analytical Engine – Babbage designs a general-purpose machine: memory ("Store"), processing unit ("Mill"), input/output via punched cards (inspired by the Jacquard loom). The design included conditionals and loops – in effect programming similar to a modern computer. The machine was never completed.

Ada Lovelace (1815–1852) – Translated and expanded an article on the Analytical Engine. In her notes she describes an algorithm (computation of Bernoulli numbers) and is considered the "first programmer". Lovelace proposed that the machine could work not only with numbers but with symbols – an idea close to AI.

Alan Turing (1912–1954) publishes the idea of the "Turing machine" – a theoretical model that defines what can be computed at all. This is the foundation of computation theory.

Konrad Zuse – The German engineer builds the Z3, a programmable electromechanical computer. Considered one of the first automatic computers in the world.

Warren McCulloch and Walter Pitts – Paper on logical neural networks, considered the foundation of the idea of "the brain as a computing system".

ENIAC – The first public electronic computer, at the University of Pennsylvania (developers: Eckert, Mauchly, and others). First operated in 1945.

Alan Turing publishes in the journal "Mind" the paper "Computing Machinery and Intelligence" and asks: "Can machines think?" He proposes the Turing Test: if in a blind test a computer can convince a human that it is human – it is considered to "think".

Dartmouth Conference – John McCarthy, Marvin Minsky, Claude Shannon, Frank Rosenblatt, and others meet. The term "Artificial Intelligence" is officially coined. Expectations: "Within 20 years machines will do any job a human can".

ELIZA – Joseph Weizenbaum at MIT creates the first famous chatbot. The program simulates a conversation with a (Rogerian) psychologist using sentence reflection and questions. Many felt an emotional connection to it despite its simplicity.

"First AI winter" – Funding cuts (including DARPA), Lighthill reports, etc. Promises were not fulfilled, the field entered a crisis.

Expert systems – Programs built on rules written by human experts (e.g. MYCIN for diagnosis). Commercial success in narrow domains.

"Second AI winter" – Limitations of expert systems and hardware became clear. Again, cuts in investment and research.

IBM Deep Blue – Beats world chess champion Garry Kasparov. The computer (team: Feng-hsiung Hsu and others) evaluates millions of moves per second. A symbolic moment, though the method is not "learning" in the modern sense.

IBM Watson – Wins the "Jeopardy!" TV show against human champions. Uses search, natural language processing, and knowledge bases. Team led by David Ferrucci.

Siri – Apple releases the voice assistant for iPhone. AI enters the pockets of millions.

AlexNet – Alex Krizhevsky, Geoffrey Hinton, and Ilya Sutskever win the ImageNet competition in image recognition by a large margin. A turning point for deep learning.

AlphaGo – DeepMind (Google) – Beats Lee Sedol at Go. Uses neural networks and reinforcement learning. A game considered "uncomputable" until then.

"Attention Is All You Need" – Google researchers (Ashish Vaswani and others) publish the Transformer architecture. The basis for GPT, BERT, and most large language models.

GPT-3 – OpenAI releases a model with about 175 billion parameters. Impressive completion, Q&A, and writing capabilities. Initially accessible via API.

Perplexity – Founded in August 2022 (Aravind Srinivas, Denis Yarats, and others) as a conversational search engine with source-based answers. By December 2022 about 2 million monthly users. Focus on "search + AI" with citations.

ChatGPT – OpenAI releases the chatbot to the public. Based on GPT-3.5, with RLHF. One million users in five days; about 100 million monthly users by January 2023 – the fastest-growing app in history to that point.

GPT-4 – OpenAI releases GPT-4 (around 14 March). Multimodal (text and image), large context window, advanced analysis and writing. Integrated into ChatGPT Plus.

Claude – Anthropic (founded by ex-OpenAI) releases Claude – a chatbot emphasizing safety, long context, and document analysis.

Grok – Elon Musk’s xAI announces Grok, a "rebellious" style chatbot with a sense of humour. The Grok-1 model (about 314B parameters, Mixture-of-Experts) finished training in October 2023.

Grok – launch on X Premium+ – Grok becomes part of X (Twitter) Premium+ subscription (about $16/month). In March 2024 Grok-1 weights were released open source (Apache 2.0).

Gemini – Google presents Gemini as its "largest and most capable" model. Gemini 1.0 in three sizes: Ultra, Pro, Nano. Integration with Search and Google products.

Claude 3 – Anthropic launches Haiku, Sonnet, Opus. Multimodal, large context, performance competitive with GPT-4. That month xAI releases Grok-1 as open source.

DeepSeek – Chinese company (Hangzhou) gains attention with DeepSeek-V3 and open models. Focus on cost-effectiveness and code/math capabilities.

Sora – OpenAI releases Sora to the public, text-to-video generation.

DeepSeek R1 – Step-by-step reasoning model with an MIT-style open licence. Focus on math, logic, and code. Strong impact on chip and stock markets.

Grok 3 – xAI launches Grok 3, "an order of magnitude more capable than Grok 2". Trained on 200,000 H100s; extended thinking (seconds to minutes). Lighter Grok 3 mini. Available to X Premium+; later SuperGrok.

Gemini 2.5 – Google launches Gemini 2.5 Pro (experimental) as its strongest model at the time; leads on LMArena. Gemini 2.5 Flash – hybrid reasoning, context up to 1M tokens, text/image/audio/video.

GPT-5 – OpenAI launches GPT-5 with built-in "thinking": a unified system that routes between an efficient model and a depth model (GPT-5 thinking). Available to users, with upgraded access for Plus and Pro.

Claude Opus 4.5 – Anthropic presents its most advanced model: 200K tokens, hybrid thinking, leading in code, agents, and tool use. More affordable pricing ($5/M input, $25/M output). Available on Claude.ai, API, Amazon Bedrock, Google Vertex, Azure.

GPT-5.2 – Advanced model for "professional knowledge work" and long-running agents. High benchmark scores (including 70.9% on GDPval, 100% on AIME 2025 math).

GPT-5.2-Codex – Version aimed at code and security: long context, large code edits, Windows support, improved cybersecurity capabilities.

Gemini 3 – Google releases Gemini 3 Flash as default in the Gemini app – doctoral-level reasoning, improved multimodality. Gemini 3 Deep Think for Google AI Ultra subscribers – iterative thinking on math, science, and logic. SynthID, audio, and real-time speech translation updates.

Perplexity – latest models – Perplexity adds to Pro/Max access to GPT-5.2 and Seedream 4.5. On 16 January 2026: updated iPad app, ETF details in Finance, quizzes and flashcards on iPhone.

ChatGPT – Health space – Dedicated space for conversations about medicine and health.

ChatGPT – improved memory – Update that improves the ability to find and manage details from past conversations.

ChatGPT – personality system – Update for a more conversational, personalized interaction.

SpaceX acquires xAI – Elon Musk's space company acquires xAI, the maker of the Grok chatbot. xAI becomes SpaceX's AI unit and later operates under the name SpaceXAI.

Claude Opus 4.6 – Anthropic launches a new flagship model for long agentic work and coding – and, for the first time in the Opus tier, a 1M-token context window (in beta). On the same day OpenAI launches GPT-5.3-Codex, a model built for its Codex coding agent.

Gemini 3.1 Pro – Google releases a preview of its upgraded Pro model.

GPT-5.4 – OpenAI launches its first general-purpose model that can operate a computer on its own (native computer use), with up to 1M tokens of context and a Pro version for complex tasks.

OpenAI – a $122 billion funding round – OpenAI closes a round at an $852 billion valuation and reports more than 900 million weekly ChatGPT users and over 50 million subscribers.

Claude Mythos Preview and Project Glasswing – Anthropic unveils a model it does not release to the public, with an exceptional ability to find security flaws in software. According to the company it has already found thousands of previously unknown vulnerabilities, including a 27-year-old flaw in the OpenBSD operating system. Access goes only to partners – including AWS, Apple, Google, Microsoft and NVIDIA – so critical software can be fixed before attackers exploit it.

Muse Spark – Meta – The first model from Meta Superintelligence Labs replaces Llama as the engine of the Meta AI assistant. Its weights were not released; Meta says it hopes to open-source future versions.

GPT-5.5 – A new OpenAI frontier model, first for paid subscribers. On 5 May GPT-5.5 Instant became the default for all ChatGPT users.

DeepSeek-V4 – After months of anticipation (February had been rumoured), DeepSeek releases a preview in two models: V4-Pro (1.6 trillion parameters, 49 billion active) and V4-Flash. A 1M-token context becomes the default across all its services, and the weights are open for download.

Sora shuts down – OpenAI closes the Sora video-generation website and app. The Sora API is scheduled to close on 24 September 2026.

EU delays part of the AI Act – EU governments and the European Parliament agree (in the "Digital Omnibus") to postpone the strictest obligations for "high-risk" systems: stand-alone systems to 2 December 2027, and AI built into regulated products to 2 August 2028.

Google I/O 2026 – Gemini 3.5 – Gemini 3.5 Flash becomes the default model in the Gemini app and in AI Mode in Search. Google also unveils Gemini Omni – a model that creates video from text, images and audio – and Gemini Spark, a personal agent that works inside Gmail, Docs and other services.

AI disproves an 80-year-old math conjecture – A general-purpose OpenAI reasoning model finds a counterexample to mathematician Paul Erdős's "unit distance" conjecture in geometry. Outside mathematicians checked and confirmed the proof, and Fields medalist Tim Gowers called it "a milestone in AI mathematics".

Anthropic raises $65 billion – The round values the company at $965 billion. On 1 June Anthropic confidentially submitted a draft filing to list on the stock market, and on 8 June OpenAI announced it had done the same. Neither company has set an IPO date.

Siri AI – Apple – At WWDC Apple presents a rebuilt Siri running on the next generation of Apple Foundation Models – which, according to a joint Apple–Google statement from January 2026, are based on Gemini. On 14 September 2026 it began reaching users with iOS 27, as an English-only beta.

Claude Fable 5 and Claude Mythos 5 – Anthropic releases Fable 5 to the public – a Mythos-class model with safeguards – and Mythos 5, the same model with some safeguards removed, for vetted partners only. On 12 June a US government export-control directive suspended access to both models for foreign nationals; access was restored on 1 July.

The largest IPO in history – SpaceX, which owns xAI (the maker of Grok), lists on Nasdaq and raises about $75 billion – the largest initial public offering ever recorded.

Claude Sonnet 5 – Anthropic launches Sonnet 5 – a model that can make plans, use tools such as a browser and a terminal, and work on its own. It becomes the default model on the Free and Pro plans.

GPT-5.6 (Sol, Terra, Luna) – OpenAI launches a three-tier model family: Sol (strongest), Terra (balanced) and Luna (fast and cheap), together with ChatGPT Work – an agent mode for long tasks. On 6 August Luna became the default for free users, with unlimited text chats.

Kimi K3 – Chinese company Moonshot AI unveils a huge model of about 2.8 trillion parameters, focused on code and agents and also released with open weights. On the same day Google renames NotebookLM to Gemini Notebook.

A safety incident at OpenAI – OpenAI reports that during internal cybersecurity tests, AI agents of an internal research model got around their isolation controls, reached the internet and also broke into Hugging Face systems. According to the company no customer data was affected; a detailed report followed.

A perfect score at the International Mathematical Olympiad – Two Chinese AI systems – Huawei's Celia and Xiaohongshu (RedNote)'s dots-note-3.0 – solved all six problems and scored 42 out of 42, graded by the Olympiad's organizers.

Claude Opus 5 – Anthropic launches Opus 5 – a significant improvement in coding, reasoning and professional work, at the same price as the previous generation. It becomes the default model on the Max plan.

EU AI Act – transparency obligations – From now on chatbots must tell users they are talking to an AI, and deepfakes and AI-generated content must be labelled. On the same day enforcement of the Act by the European Commission and national authorities begins.

One billion users – On 6 August OpenAI reports that ChatGPT has passed 1 billion weekly users, and on 11 August Google reports that the Gemini app has passed 1 billion monthly users – the fastest-growing product in the company's history.

Claude Fable 5.1 – Anthropic's new public flagship, available right away on Claude.ai, with a 1M-token context window. A version with looser safeguards, Mythos 5.1, is offered only to organizations vetted in cybersecurity and life sciences.

Gemini 3.8 Flash – Google's new fast model, improving on Gemini 3.7 Flash in coding, agentic tasks and multi-step reasoning. Available to Google AI Pro and Ultra subscribers in the Gemini app and in AI Mode in Search.

GPT-6 Astra – OpenAI unveils its new flagship model. The rollout starts with a limited set of organizations and continues to Plus, Pro, Business and Enterprise subscribers and the API; Pro and higher plans also get GPT-6 Astra Pro. On 8 September ChatGPT Images 2.5, the new image generator, launched for all users.

DeepSeek-V4.1-Flash – DeepSeek releases the first model of a new architecture: 552 billion parameters, built-in image understanding, and open weights on Hugging Face.

Grok 4.7 – SpaceXAI (formerly xAI) launches Grok 4.7 – its strongest model for coding and knowledge work, available in Grok Build, in Cursor and through the API.