What is an LLM?
An LLM (Large Language Model) is a neural network trained on huge amounts of text to predict the next word (token) in a sentence. Once the model is large enough and trained on enough data, it can write, translate, summarise, program and reason. Almost every modern LLM is built on the transformer architecture that Google researchers described in 2017.
AI (artificial intelligence) is the broader term: computer systems that solve tasks that normally require human intelligence. LLMs are just one – currently the most visible – branch of AI.
Who invented AI?
There is no single inventor, but two names keep coming up:
- Alan Turing (UK) asked "Can machines think?" in 1950 and proposed the Turing test.
- John McCarthy (US) coined the term artificial intelligence in 1955 and in 1956 gathered researchers for the Dartmouth workshop together with Marvin Minsky, Claude Shannon and Nathaniel Rochester. It is regarded as the birth of AI research.
Original model or fine-tune?
Training a language model from scratch (pre-training) is expensive: it takes thousands of GPUs, trillions of tokens and months of work. Many "national" AI models are therefore fine-tuned versions of open models such as Meta's Llama or France's Mistral – an existing model trained further on, say, local languages. That is a fine and legitimate technique, but it is not the same as building your own model.
United States
OpenAI – the GPT series
- Models: GPT-3, GPT-4, o1, GPT-5 (ChatGPT)
- Founded: December 2015 in San Francisco
- Founders: including Sam Altman, Greg Brockman and Ilya Sutskever
- CEO: Sam Altman
Google – the Gemini and PaLM series
- Models: PaLM, Gemini (and the open Gemma models)
- Founded: Google in 1998 by Larry Page and Sergey Brin; the AI lab DeepMind in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman (acquired by Google in 2014)
- CEO: Sundar Pichai (Google/Alphabet) – Demis Hassabis runs Google DeepMind
Meta – the Llama series
- Models: Llama 1–4 (open weights); in 2026 the closed Muse Spark followed
- Founded: Facebook in 2004 by Mark Zuckerberg and others; the FAIR research lab in 2013
- CEO: Mark Zuckerberg – Alexandr Wang leads Meta Superintelligence Labs
Anthropic – the Claude series
- Models: Claude (Haiku, Sonnet, Opus and Fable) – see also Claude AI models compared
- Founded: 2021 in San Francisco
- Founders: siblings Dario and Daniela Amodei together with other former OpenAI staff
- CEO: Dario Amodei
xAI – the Grok series
- Models: Grok
- Founded: 2023 by Elon Musk
- CEO: Elon Musk – xAI was acquired by SpaceX in February 2026 and has since been folded into SpaceX as the SpaceXAI division
Amazon – the Titan and Nova models
- Models: Titan (2023), succeeded by Nova (December 2024)
- Founded: Amazon in 1994 by Jeff Bezos
- CEO: Andy Jassy
India
India has many AI projects, but only a few language models that are actually trained from scratch.
Sarvam AI
- Models: Sarvam 30B and Sarvam 105B (February 2026), trained from scratch on Indian infrastructure and supporting all 22 official Indian languages. The earlier Sarvam-M (2025), by contrast, was a fine-tuned Mistral model.
- Founded: 2023 in Bengaluru
- Founders: Vivek Raghavan and Pratyush Kumar
Krutrim
- Models: Krutrim 1 and Krutrim 2
- Founded: 2023
- Founder/leader: Bhavish Aggarwal (also founder of Ola)
Hanooman and BharatGPT
Hanooman (Seetha Mahalaxmi Healthcare with the IIT Bombay-led BharatGPT consortium) and CoRover's BharatGPT are marketed as Indian language models, but have been criticised for largely building on open models such as Llama and Mistral rather than being trained from scratch.
Europe – more than just Mistral
Mistral AI (France)
- Models: Mistral, Mixtral, Mistral Large, Le Chat
- Founded: April 2023 in Paris
- Founders: Arthur Mensch, Guillaume Lample and Timothée Lacroix
- CEO: Arthur Mensch
Aleph Alpha (Germany)
- Models: Luminous, trained from scratch with a strong focus on European law, compliance (GDPR) and public-sector data. In October 2026 they launched Kolibri, a "sovereign" model for government and industry.
- Founded: 2019 in Heidelberg
- Founders: Jonas Andrulis and Samuel Weinbach
- Leadership: Andrulis has stepped down; Ilhan Scheer is CEO. In 2026 Aleph Alpha agreed to merge with Canada's Cohere.
Silo AI (Finland)
- Models: the open Poro and Viking models, built from scratch with the University of Turku (TurkuNLP) and trained on LUMI, one of Europe's most powerful supercomputers. The goal is models that understand the Nordic and European languages – including the smaller ones.
- Founded: 2017 in Helsinki
- Founder/CEO: Peter Sarlin
- Owner: acquired by AMD in 2024 for about USD 665 million
Kyutai (France)
- Models: Moshi, a fully original open-source model built from scratch and specialised in real-time spoken conversation – it listens and talks at the same time, like a real conversation.
- Founded: November 2023 in Paris as a non-profit AI lab, independent of Mistral
- Founders/backers: Xavier Niel (Iliad/Free), Rodolphe Saadé (CMA CGM) and Eric Schmidt (former Google CEO)