The AI Timeline
From the Transformer paper to today. Every major model release, company event, and industry milestone.
AI API Prices Drop 90% from 2023 Levels
By early 2026, API pricing for frontier models has dropped approximately 90% from 2023 levels. GPT-4o-class intelligence now costs less than GPT-3.5 did at launch. Budget models like Gemini Flash and GPT-4o-mini cost under $0.50 per million tokens, making AI integration viable for virtually any application.
AI Job Displacement Debate Intensifies
Multiple studies from McKinsey, Goldman Sachs, and the World Economic Forum confirm AI is reshaping white-collar employment. Early data shows mixed results: customer support and data entry roles decline, but AI trainer, prompt engineer, and AI integration roles grow faster. The consensus shifts from "AI will take your job" to "someone using AI will take your job."
Anthropic Launches Claude Opus 4 — Tops the Leaderboard
Anthropic launches Claude Opus 4 with the highest Arena Elo score of any model (1380). It takes the overall #1 position on the LMSYS Chatbot Arena leaderboard, with particular dominance in long-form writing, analysis, and coding tasks.
Read more →Alibaba Releases Qwen 3 — Chinese AI Goes Global
Alibaba releases the Qwen 3 model family, including open-source versions up to 72B parameters. The 235B flagship model matches GPT-4o on several benchmarks while excelling at Chinese, Japanese, and Korean language tasks. It signals that AI leadership is no longer exclusively a US phenomenon.
Read more →OpenAI Releases o3 — Near-Perfect Reasoning
OpenAI releases o3, its most powerful reasoning model. It scores 96.2% on HumanEval (coding) and 78.1% on GPQA (PhD-level science), approaching expert-level performance on the hardest AI benchmarks. The model "thinks" for minutes before answering complex questions.
Read more →Meta Releases Llama 4 — Open Source Goes Multimodal
Meta releases Llama 4, including the Maverick and Behemoth variants. With a 1 million token context window and native multimodal support, open-source models definitively reach frontier quality. Llama 4 Maverick becomes the most cost-effective quality model via API providers.
Read more →Google Launches Gemini 2.5 Pro
Google launches Gemini 2.5 Pro with significantly improved reasoning capabilities while maintaining its industry-leading 1 million token context window. It closes the gap with Claude and GPT on quality benchmarks while remaining the context window champion.
Read more →NVIDIA B200 GPUs Begin Shipping — Next-Gen AI Compute
NVIDIA begins volume shipments of B200 Blackwell GPUs, offering 2.5x the training performance and 5x the inference performance of the H100. Pricing starts at approximately $30,000-40,000 per unit, and every major cloud provider races to build Blackwell clusters.
xAI Launches Grok 3 — Elon's AI Contender
Elon Musk's xAI launches Grok 3, trained on the Colossus supercomputer with 100,000 NVIDIA H100 GPUs — the largest training cluster ever assembled. It competes with GPT-4o and Claude Sonnet on benchmarks and integrates real-time X (Twitter) data.
Read more →DeepSeek R1 Open-Sources Chain-of-Thought Reasoning
DeepSeek releases R1, an open-source reasoning model that matches OpenAI's o1 on math and coding benchmarks. It sends tech stocks tumbling and forces Western AI companies to reconsider their cost assumptions. Open-source reasoning models are now a reality.
Claude 3.5 Sonnet Becomes the Developer Favorite
Claude 3.5 Sonnet solidifies its position as the most popular model among developers. It tops multiple developer preference surveys, particularly for coding tasks, and becomes the default model in Cursor, Windsurf, and other AI coding tools.
Perplexity Hits 100M Monthly Users
Perplexity reaches 100 million monthly active users, establishing itself as the first serious challenger to Google Search in over a decade. Its AI-powered answers with cited sources resonate with users frustrated by ad-heavy search results and SEO spam.
DeepSeek V3 Matches GPT-4o at Fraction of Cost
Chinese AI lab DeepSeek releases V3, an open-source model that matches GPT-4o on key benchmarks while reportedly training for just $5.5 million — a fraction of the hundreds of millions spent by Western labs. It challenges the assumption that frontier AI requires massive budgets.
OpenAI Restructures — For-Profit Transition Begins
OpenAI raises $6.6 billion at a $157 billion valuation and begins restructuring from its original nonprofit model to a capped-profit corporation. The move triggers criticism from co-founder Elon Musk (who sues) and raises questions about AI safety organizations prioritizing growth over mission.
OpenAI Launches o1 — Chain-of-Thought Reasoning
OpenAI launches o1 (codename "Strawberry"), a model that uses chain-of-thought reasoning — visible "thinking" before answering. It dramatically outperforms GPT-4o on math competitions, coding challenges, and PhD-level science questions, but is slower and more expensive.
Read more →Cursor Raises $60M, AI Coding Goes Mainstream
Cursor raises $60M at a $400M+ valuation as AI-native code editors challenge VS Code's dominance. The success of Cursor, alongside GitHub Copilot surpassing 1.8M paid subscribers, confirms that AI coding tools are now essential developer infrastructure.
DALL-E 3 Becomes Free in ChatGPT
OpenAI makes DALL-E 3 image generation available to free ChatGPT users, removing the last major barrier to AI image creation. Previously locked behind the $20/month Plus subscription, the move pressures competitors like Midjourney to reconsider their pricing.
OpenAI Launches GPT-4o — Free Multimodal AI
OpenAI launches GPT-4o ("omni"), a natively multimodal model that processes text, vision, and audio in a single architecture. For the first time, free ChatGPT users get GPT-4 level intelligence. The new voice mode demos stun the public.
Read more →Meta Releases Llama 3 — Open Source Catches Up
Meta releases Llama 3 with 8B and 70B parameter versions, dramatically closing the gap between open-source and proprietary models. The 70B model matches GPT-3.5 on most benchmarks. The later 405B version approaches GPT-4 level.
Amazon Invests $4B in Anthropic
Amazon completes its $4 billion investment in Anthropic, the largest AI investment outside of the Microsoft-OpenAI partnership. Anthropic commits to using AWS as its primary cloud provider, and Claude becomes a first-class model on Amazon Bedrock.
Anthropic Launches Claude 3 Family
Anthropic launches the Claude 3 family: Haiku (fast/cheap), Sonnet (balanced), and Opus (flagship). Claude 3 Opus matches or exceeds GPT-4 on most benchmarks, establishing Anthropic as a true competitor to OpenAI for the first time.
Mistral Releases Mistral Large — Europe's AI Champion
French AI startup Mistral releases Mistral Large, its flagship model competitive with GPT-4 on key benchmarks. Backed by European investors and positioned as the EU-compliant alternative to American AI models.
Read more →NVIDIA Passes $2 Trillion Market Cap
NVIDIA passes $2 trillion market cap, making it one of the most valuable companies on Earth. AI chip demand drives record revenue of $22 billion in a single quarter. Every major tech company is buying as many GPUs as NVIDIA can produce.
Google Launches Gemini 1.5 Pro with 1M Context
Google launches Gemini 1.5 Pro with a 1 million token context window — roughly 2,500 pages of text. It's 10x larger than any competitor and opens new use cases: analyzing entire codebases, books, and video transcripts in a single prompt.
Read more →OpenAI Unveils Sora — Text to Video
OpenAI unveils Sora, a text-to-video model that generates photorealistic 60-second clips from text descriptions. The demo videos — a woman walking through Tokyo, woolly mammoths in snow — stun the creative industry and spark debates about the future of video production.
ElevenLabs Raises $80M at $1B+ Valuation
ElevenLabs raises $80 million at a $1.1 billion valuation, making it a unicorn just two years after founding. The funding validates voice AI as a major market category and accelerates development of real-time voice translation and dubbing features.
EU Passes the AI Act — First Major AI Regulation
The European Union passes the AI Act, the world's first comprehensive legislation governing artificial intelligence. It establishes a risk-based framework: banned AI uses (social scoring), high-risk requirements (transparency, human oversight), and foundation model rules.
Sam Altman Fired and Rehired at OpenAI
OpenAI's board fires CEO Sam Altman over a vague loss of confidence. Within days, 95% of staff threaten to quit, Microsoft offers to hire the entire team, and Altman is reinstated. The crisis reveals the tension between OpenAI's nonprofit mission and commercial ambitions.
OpenAI Launches GPT-4 Turbo — Cheaper, Faster, 128K Context
OpenAI launches GPT-4 Turbo at its first DevDay conference. Key upgrades: 128K context window (4x the original GPT-4), 3x cheaper pricing, JSON mode for structured output, and knowledge updated to April 2023. The 128K context window becomes the new industry standard.
Biden Signs Executive Order on AI Safety
President Biden signs the Executive Order on Safe, Secure, and Trustworthy AI — the most comprehensive US government action on artificial intelligence. It requires companies to share safety test results with the government before releasing powerful AI models and establishes new standards for AI safety.
OpenAI Cuts GPT-4 API Prices by 50%
OpenAI cuts GPT-4 API prices by approximately 50%, signaling the beginning of an aggressive pricing war. The move forces Anthropic and Google to match, establishing the trend of rapidly falling AI costs that continues through 2026.
NVIDIA H100 GPUs Sell for $40,000+ Amid AI Boom
NVIDIA H100 GPUs sell for $30,000-40,000+ each as demand from AI companies far exceeds supply. NVIDIA's market cap surpasses $1 trillion, making Jensen Huang the face of the AI hardware boom. Cloud GPU availability becomes a strategic resource.
Anthropic Raises $750M Series C
Anthropic raises $750 million in Series C funding led by Spark Capital, valuing the company at $4.1 billion. The funding fuels Claude model development and positions Anthropic as the primary competitor to OpenAI in the foundation model market.
Google Launches Bard (Later Renamed Gemini)
Google rushes to launch Bard, its ChatGPT competitor, powered by LaMDA and later PaLM 2. The launch is rocky — a factual error in the demo costs Google $100 billion in market cap. Bard is later rebranded as Gemini.
Midjourney V5 Sets New Quality Standard
Midjourney releases V5, a massive leap in image quality. Photorealistic human faces, accurate hands, and detailed textures make AI-generated images nearly indistinguishable from photographs for the first time.
Anthropic Launches Claude — The Safety-First AI
Anthropic launches Claude, trained using Constitutional AI (RLHF with a set of written principles). It positions itself as the safety-focused alternative to ChatGPT, with a focus on being helpful, harmless, and honest.
GPT-4 Launches — Multimodal AI Arrives
OpenAI launches GPT-4, its first multimodal model that can understand images. It passes the bar exam in the 90th percentile, scores a 5 on AP exams, and marks a massive leap in reasoning capability over GPT-3.5.
Read more →Meta Releases Llama — Open Source Shakes Up AI
Meta releases Llama (Large Language Model Meta AI) as open source. Smaller than GPT-4 but remarkably capable, it proves that open-source models can compete with proprietary ones and spawns an explosion of fine-tuned variants.
Microsoft Launches Bing AI with GPT-4
Microsoft integrates GPT-4 into Bing Search, creating the first AI-powered search engine from a major tech company. The chatbot goes viral for its sometimes unhinged responses, generating massive media coverage.
ChatGPT Launches — AI Goes Mainstream
OpenAI launches ChatGPT based on GPT-3.5. It reaches 100 million users in just two months — the fastest-growing consumer application in history. AI becomes a dinner-table conversation topic overnight.
DALL-E 2 Opens to the Public
OpenAI opens DALL-E 2 to the public without a waitlist. For the first time, anyone can type a text description and receive AI-generated images. Combined with Midjourney and Stable Diffusion's releases, 2022 becomes the year AI image generation goes mainstream.
Stable Diffusion Released as Open Source
Stability AI releases Stable Diffusion as open source. For the first time, anyone with a decent GPU can generate AI images locally for free. The release spawns thousands of fine-tuned models, LoRAs, and a massive open-source ecosystem.
GitHub Copilot Launches in Preview
GitHub launches Copilot as a technical preview, powered by OpenAI Codex. It suggests code completions directly in VS Code and becomes the first AI coding tool to achieve widespread adoption. Within two years, it surpasses 1 million paying subscribers.
NVIDIA A100 GPUs Become the Standard for AI Training
The NVIDIA A100 (80GB) becomes the gold standard for AI training. Cloud providers race to stockpile them, creating a GPU shortage that drives wait times to 6+ months and establishes NVIDIA's monopoly on AI compute.
Anthropic Founded by Former OpenAI Researchers
Dario Amodei, Daniela Amodei, and several other researchers leave OpenAI to found Anthropic, with a focus on AI safety research. The company will go on to create the Claude model family.
OpenAI Unveils DALL-E — Text to Images
OpenAI unveils DALL-E, a GPT-3 variant that generates images from text descriptions. "An armchair in the shape of an avocado" becomes the iconic demo, marking the birth of the AI image generation era.
Microsoft Invests $1B in OpenAI
Microsoft invests $1 billion in OpenAI, securing exclusive cloud computing rights and laying the groundwork for Azure OpenAI Service. This partnership will eventually integrate GPT into Bing, Office 365, and GitHub Copilot.
OpenAI Launches GPT-3 — The AI Revolution Begins
OpenAI launches GPT-3 with 175 billion parameters — 100x larger than GPT-2. Its ability to write essays, code, and poetry with just a few examples (few-shot learning) stuns the AI community and captures mainstream attention for the first time.
OpenAI Releases GPT-2 — 'Too Dangerous to Release'
OpenAI releases GPT-2 with 1.5 billion parameters and initially withholds the full model, claiming it's "too dangerous to release" due to potential misuse for generating disinformation. The decision sparks the first major debate about AI safety and open access.
Google Releases BERT
Google releases BERT (Bidirectional Encoder Representations from Transformers). It becomes the backbone of Google Search and sets new records on 11 NLP benchmarks simultaneously.
OpenAI Releases GPT-1
OpenAI releases GPT-1 with 117 million parameters. It demonstrates that pre-training a Transformer on a large corpus of text, then fine-tuning on specific tasks, produces surprisingly capable results.
Google Publishes 'Attention Is All You Need'
Google Brain researchers publish "Attention Is All You Need," introducing the Transformer architecture. This single paper becomes the foundation for GPT, BERT, Claude, Gemini, and every other modern language model.