Skip to content
BENIBIX Learn

OpenAI, Anthropic and Google

Level 1 · The AI labs

The three US labs behind most closed AI models. Who makes GPT, Claude and Gemini, when each began, and how you reach them.

Course contentsLesson 2 of 13

After this lesson you can

  • Match GPT, Claude and Gemini to their labs
  • Name the size tiers in the Claude family
  • List the three ways to reach a closed model

11 mincovers 6 glossary entries

Worth reading first: AI Labs and Model Families

Who makes GPT, Claude and Gemini

Three US labs make most of the closed models people use. OpenAI makes GPT and the ChatGPT app, which came out on 30 November 2022. Anthropic makes Claude, first released in March 2023. Google DeepMind makes Gemini, first released in December 2023. All three sell their models through an app and an API.

OpenAI began in San Francisco in December 2015, as a nonprofit lab. Its GPT models grew with each version. GPT-3, in 2020, showed what a very large model could do. ChatGPT made the idea famous. OpenAI also runs a line of models that think before they answer, and it makes picture and speech models.

Anthropic was founded in 2021 by people who left OpenAI. It is a public benefit company, so its rules name a public goal beside profit. It is known for its work on making models safe. Its Claude models come in tiers. Claude Haiku is the fastest. Then come Claude Sonnet and Claude Opus, and Claude Fable is the strongest.

Google DeepMind is Google's main AI lab. DeepMind began in London in 2010, Google bought it in 2014, and it merged with Google Brain in 2023. Its Gemini models were built from the start to take in text, pictures, sound and video. They run inside Google's own products and the Gemini app.

All three keep their main models closed. You cannot download GPT, Claude or Gemini. You reach them in one of three ways: the lab's app, the lab's API, or a big cloud. Claude, for one, is also sold through Amazon Bedrock and Google's cloud, so a firm can use the cloud it already pays for.

From the outside the three look alike: a chat box, a paid plan and an API. The difference is in the details. Each family has its own sizes, prices and strengths, and the lead changes hands often. A test you ran six months ago may not hold today.

People often think the app and the model are one thing. The same Claude model sits behind the Claude app, Claude Code and many apps built by other firms. And the Gemini app is one product among many that run on Gemini models. Pick the model for the job, then the app.

The three big closed labs

LabBeganModel familyChat app
OpenAI2015GPTChatGPT
Anthropic2021ClaudeClaude app
Google DeepMind2010, as DeepMindGeminiGemini app
The GPT family grew for four years before ChatGPT put it in front of the public.

Worked example

Jack's firm uses all three

Jack's firm sends about 3,000 requests a day to AI models. It already pays Google for its cloud, so it reaches Gemini and Claude there, on one bill.

One task, sorting photos of receipts, works best on a GPT model. For that one job it calls OpenAI's own API, about 400 times a day.

The words for this lesson6

Check yourself

Who makes the Claude models?

Finished this lesson?

Marking it complete also marks its 6 glossary entries learned, which updates your roadmap and your saved page. 0 are already marked.