The AI landscape

ChatGPT launches

OpenAI opens a free conversational assistant to the public. It reaches a mass audience within weeks and introduces millions of people to large language models.

OpenAI

Entry: LLM (large language model)Entry: Prompt

Bing Chat, Bard and LLaMA

Microsoft adds an assistant to its search engine, Google announces Bard and Meta shares LLaMA with researchers: the competition begins.

Microsoft, Google, Meta

GPT-4 and the first Claude

OpenAI unveils GPT-4, much more capable and able to read images. Anthropic launches Claude, its own assistant.

OpenAI, Anthropic

Entry: LLM (large language model)

Llama 2, an open model

Meta releases Llama 2 with a licence allowing commercial use: anyone can download the model and run it on their own machines.

Meta

Mistral AI enters the scene

The young French start-up releases Mistral 7B under a free licence, then Mixtral in December. Europe now has a recognised player.

Mistral AI

First global AI safety summit

At Bletchley Park in the UK, around thirty countries sign a joint declaration on the risks of the most advanced models.

Governments

Gemini and a deal on the AI Act

Google unveils Gemini, designed from the start for text, images and sound. The same month, the European Union reaches a political agreement on the AI Act, the first major AI law.

Google, European Union

Sora shows generated video

OpenAI unveils Sora, able to produce realistic videos from a simple description. Bard is renamed Gemini.

OpenAI, Google

GPT-4o, the assistant that talks

Near-instant voice conversation and live image understanding: the assistant becomes a natural conversation partner.

OpenAI

AI in your phone

Apple announces Apple Intelligence. Nvidia, which makes the chips AI depends on, briefly becomes the world’s most valuable company.

Apple, Nvidia

The AI Act comes into force

The European regulation applies gradually: some uses are banned from 2025, obligations for general-purpose models follow.

European Union

Models that “think”

With o1, OpenAI launches reasoning models: they take time to break a problem down before answering, with big gains in maths and code.

OpenAI

Two Nobel prizes for AI

The Nobel Prize in Physics goes to John Hopfield and Geoffrey Hinton (neural networks), the Chemistry prize to Demis Hassabis and John Jumper (protein structure prediction with AlphaFold) and David Baker (computational protein design).

Research

A standard to plug AI into your tools

Anthropic publishes the Model Context Protocol (MCP), an open standard for connecting an assistant to files, software and services. It is quickly adopted across the ecosystem.

Anthropic

Entry: AI agent

The DeepSeek shock

Chinese start-up DeepSeek releases R1, an open reasoning model close to the best, reportedly much cheaper to train. Tech stocks wobble.

DeepSeek

Paris AI Action Summit

France hosts an international summit on AI uses, investment and governance. The first AI Act bans take effect.

Governments

AI that codes for you

Anthropic launches Claude Code, an agent that works directly in developers’ projects. Coding assistants become mainstream.

Anthropic

Entry: AI agent

Image generation for everyone

Image creation built into ChatGPT sparks a worldwide viral trend; Google releases Gemini 2.5.

OpenAI, Google

Video with sound and Claude 4

Google unveils Veo 3, which generates video and sound together. Anthropic launches Claude 4, built for long autonomous tasks.

Google, Anthropic

GPT-5 and European obligations

OpenAI launches GPT-5. In Europe, transparency obligations for providers of general-purpose models start to apply.

OpenAI, European Union

Gemini 3

Google launches a new generation focused on reasoning and agentic coding, with a deep thinking mode (Deep Think).

Google

Mistral Large 3, a large European open model

Mistral releases a very large downloadable model that can read images and becomes its benchmark for 2026.

Mistral AI

AI that uses the computer becomes the norm

GPT-5.4 is OpenAI’s first mainline model able to click, type and browse on its own; Google builds the same ability into Gemini 3.5 Flash in June.

OpenAI, Google

Entry: AI agent

DeepSeek V4 open, Meta closes up

DeepSeek releases V4 under a free licence. Meta, by contrast, replaces its open Llama models with a closed range, Muse.

DeepSeek, Meta

A tier above the flagships

After a Claude Mythos preview reserved for a few partners in April, Anthropic opens its Mythos tier, above Opus: Claude Fable 5 is open to the public with stronger safeguards, Claude Mythos 5 reserved for restricted access.

Anthropic

GPT-6

OpenAI launches the GPT-6 generation in three models: Astra (most capable), Sol (coding) and Luna (budget).

OpenAI

A model held back for safety reasons

According to the Wall Street Journal, OpenAI decides not to release GPT-6.1 Astra, judged insufficient in safety testing; Google first reserves Gemini 4 for cyber-defence teams.

OpenAI, Google