Where did each AI come from? Who built it, when, and what is it related to? This is the complete map of the AI landscape as of April 2026 — from the foundational research to the products you use today.
Every AI product you can use today sits at the end of a chain that leads back to the same foundational research. Here’s how the key relationships work.
GPT-3 (2020) was the model that showed the world what language AI could do. GPT-4 (2023) was the first genuinely multimodal model — it could understand images as well as text. GPT-4o (2024) added native audio and became the engine inside ChatGPT. Microsoft Copilot runs on GPT-4o via a partnership with OpenAI.
BERT (2018) transformed how Google understood search queries. PaLM (2022) and PaLM 2 were the precursors to Gemini. Gemini 1.0 launched in late 2023 as Google’s answer to GPT-4. Gemini 1.5 introduced 1M+ token context windows. Gemini 2.0 is the current production model, deeply integrated into Google Workspace.
Anthropic was founded in 2021 by Dario Amodei, Daniela Amodei, and others who left OpenAI. Their safety-focused approach — Constitutional AI — trains Claude to critique and improve its own outputs. Claude 3 (2024) introduced Haiku, Sonnet, and Opus tiers. Claude 3.5 set new benchmarks for coding and reasoning. Claude 4 (2025–26) is the current frontier model.
Meta’s Llama models are open-weight — you can download and run them yourself. Llama 3 (2024) matched the performance of many commercial models. Mistral AI (Paris, 2023) produces efficient open-weight models with strong multilingual performance. Both underlie hundreds of third-party AI applications and custom deployments.