# Large language models: Overview

Source page: https://www.materialinsights.org/technologies/large_language_models/tech-overview/

#### What it does

The job the technology does for people, in a short paragraph

A large language model writes and reads text. Given a question or an instruction, it writes an answer, a summary, a translation or a program. The transformer design behind it dates from 2017 ^[1], paid access began in 2020 ^[2], and by October 2025 ChatGPT alone had 800 million weekly users, by OpenAI's own count ^[3].

as of 2026-11-11

#### How it works

The working principle in plain words, step by step

Text is first cut into tokens, short pieces of words, and each token becomes a list of numbers. A transformer network then passes these lists through many layers. In each layer, self-attention lets every token look at all the others and weigh which matter for its meaning ^[1]. During training the model reads vast amounts of text and, at each position, tries to predict the next token; backpropagation adjusts its very many weights to shrink each error. To predict well, it has to pick up grammar, facts and common patterns of argument. Later training on examples and human ratings teaches it to follow instructions. When used, it writes one token at a time, each time choosing a likely next token given everything so far.

as of 2026-11-11

#### Key figures

The handful of numbers that size it up, each with its unit, year and source

| Figure | Value | Unit | Year | Source |
|---|---|---|---|---|
| Weekly active users of ChatGPT | 800 | million | 2025 | ^[3] |
| Tokens processed per minute on OpenAI's API | 6 | billion (over) | 2025 | ^[3] |
| Developers who have built with OpenAI | 4 | million | 2025 | ^[3] |
| Cost of GPT-3.5-level output, November 2022 | 20 | USD per million tokens | 2022 | ^[4] |
| Cost of GPT-3.5-level output, October 2024 | 0.07 | USD per million tokens | 2024 | ^[4] |

as of 2026-11-11. User, developer and token counts are OpenAI's own, given by its chief executive at its developer day and reported by TechCrunch; the costs are from the Stanford AI Index for models scoring like GPT-3.5 on the MMLU test.

#### Where it is used

The main places and products the technology is found in

- Chat assistants for questions, writing and study
- Programming tools that write and explain code
- AI summaries in web search results ^[9]
- Customer service, translation and office software
- Research tools that read and summarise papers

as of 2026-11-11

#### Scale today

How much of it there is now: units in use, made each year and who uses it

**Made per year**
: Over 6 billion tokens a minute processed on OpenAI's API in October 2025 ^[3]

**Users**
: 800 million weekly users of ChatGPT alone in October 2025 ^[3]

**Unit**
: users; tokens

**Year**
: 2025

**Note**
: Counts one company's figures; other labs' models add more.

as of 2026-11-11

#### Sources

1. [Transformer: A Novel Neural Network Architecture for Language Understanding](https://research.google/blog/transformer-a-novel-neural-network-architecture-for-language-understanding/) Google Research Blog (Jakob Uszkoreit), 2017
2. [OpenAI API](https://openai.com/index/openai-api/) OpenAI, 2020
3. [Sam Altman says ChatGPT has hit 800M weekly active users](https://techcrunch.com/2025/10/06/sam-altman-says-chatgpt-has-hit-800m-weekly-active-users/) TechCrunch (Rebecca Bellan), 2025
4. [AI Index 2025: State of AI in 10 Charts](https://hai.stanford.edu/news/ai-index-2025-state-of-ai-in-10-charts) Stanford Institute for Human-Centered AI (HAI), 2025
5. [Alphabet Announces Fourth Quarter and Fiscal Year 2024 Results (Exhibit 99.1)](https://www.sec.gov/Archives/edgar/data/1652044/000165204425000010/googexhibit991q42024.htm) Alphabet Inc., filed with the US SEC, 2025

## Other sections

- [Summary](https://www.materialinsights.org/technologies/large_language_models/index.md)
- [Science](https://www.materialinsights.org/technologies/large_language_models/tech-science/index.md)
- [Evolution](https://www.materialinsights.org/technologies/large_language_models/tech-evolution/index.md)
- [Energy footprint](https://www.materialinsights.org/technologies/large_language_models/tech-energy-footprint/index.md)
- [Industries](https://www.materialinsights.org/technologies/large_language_models/tech-industries/index.md)
