If you read AI news for more than a week, you’ll run into names that sound like a cross between a star chart and a poetry class: Astra, Sol, Luna, Fable, Opus, Sonnet, Haiku, Flash. Add a version number like 5.5 or 3.8, and it’s easy to feel lost.

The good news: once you know a few basic ideas, those names start to make sense. This guide explains them.

“AI” isn’t one thing

People often talk about “AI” as if it were a single machine somewhere. It’s more like “cars.” Many different companies build them, each has several models, and new ones come out all the time.

The companies that build the most advanced AI are often called AI labs. The ones you’ll see in the news most:

  • OpenAI makes ChatGPT. Microsoft is a major investor.
  • Anthropic makes Claude. Amazon and Google are major investors.
  • Google makes Gemini, built by its research lab, Google DeepMind.
  • Meta (Facebook, Instagram, WhatsApp) makes Meta AI and the new Muse assistant.
  • xAI, Elon Musk’s AI company, makes Grok.
  • Microsoft makes Copilot, which uses a mix of Microsoft’s own AI and AI from OpenAI and Anthropic.

Some companies build their assistants on other companies’ AI. Apple’s newest Siri, for example, is built on technology from Google’s Gemini. Amazon’s Alexa+ uses Amazon’s own AI along with Anthropic’s.

There are also well-known labs outside the U.S., including Mistral in France and, in China, DeepSeek, Alibaba (maker of Qwen), and Moonshot AI (maker of the Kimi chatbot). Chinese labs often let anyone download their models for free; more on that below.

The app and the engine

Here’s the single most useful idea in this guide.

The program you open, such as ChatGPT, Claude, or Gemini, is the app. The AI system doing the thinking behind it is called the model. Think of the app as the car and the model as the engine.

That’s why you’ll hear that “ChatGPT got an upgrade” even though the ChatGPT app looks the same. OpenAI swapped in a new engine. It also explains why:

  • One app can offer several engines. Paid plans often let you pick a more powerful model for harder questions.
  • Free and paid versions can behave differently. Free plans usually run a smaller, faster model. The biggest models typically require a paid plan, often around $20 a month.
  • Different apps can share an engine. Microsoft’s Copilot and Apple’s Siri both use models made by other companies.

When a news story mentions a name like “GPT-6 Sol” or “Claude Opus 5.5,” it’s almost always talking about an engine, not an app.

How to decode a model name

Most model names have three parts:

  1. The family, which tells you the company: GPT is OpenAI, Claude is Anthropic, Gemini is Google, Grok is xAI, Muse is Meta.
  2. A version number. Higher usually means newer. A jump from 5 to 6 is a big new generation; 5.5 to 5.6 is a smaller step.
  3. A tier name, which tells you the size. Bigger models are more capable but slower and more expensive to run. Smaller ones are quicker and cheaper, and often plenty good for everyday questions.

So GPT-6 Sol means: OpenAI’s GPT family, generation 6, the “Sol” tier.

Each company picks its own theme for tier names:

  • OpenAI uses Latin sky words. From biggest to smallest, Astra (stars), Sol (sun), and Luna (moon). Its previous generation also had a middle tier, Terra (earth). OpenAI says the number marks the generation and the names mark “durable capability tiers.”
  • Anthropic uses literary forms. A haiku is a short poem, a sonnet is longer, and an opus is a major work. Its newest top tier is Fable, which Anthropic says comes from the Latin fabula, “that which is told.”
  • Google uses plain descriptions: Pro for the most capable, Flash for fast, and Flash-Lite for fastest and cheapest.

Model name decoder

Here’s who’s who right now, from most capable to fastest.

Current as of September 27, 2026. New models arrive often; we update this list as they do.

OpenAI · app: ChatGPT

  • Most capable: GPT-6 Astra
  • Everyday: GPT-6 Sol
  • Fastest and cheapest: GPT-6 Luna

Tier names are Latin for stars, sun, and moon. The earlier GPT-5.6 generation also had a middle tier, Terra (earth).

Anthropic · app: Claude

  • Most capable: Claude Fable 5.1
  • Everyday: Claude Opus 5.5 and Claude Sonnet 5
  • Fastest and cheapest: Claude Haiku 4.5

Opus 5.5 is the first of a new "5.5" generation; Sonnet 5.5 and Haiku 5.5 are expected soon.

Google · app: Gemini

  • Most capable: Gemini 3.1 Pro
  • Everyday: Gemini 3.8 Flash
  • Fastest and cheapest: Gemini 3.5 Flash-Lite

Built by Google DeepMind. Flash has been updated every few weeks; Pro was last updated in February.

Meta · app: Meta AI and Muse

  • Current model: Muse Spark

One main model for now. Meta's older Llama models are free to download.

xAI · app: Grok

  • Current model: Grok 4.7

Available in the Grok app, on X, and in Tesla cars.

Moonshot AI · app: Kimi

  • Current model: Kimi K3

Based in Beijing and backed by Alibaba and Tencent. Kimi K3 (July 2026) is an open-weight model, so anyone can download it.

Microsoft · app: Copilot

    Copilot uses Microsoft's own models along with models from OpenAI and Anthropic, depending on the task and the version of Copilot.

    What “frontier” means

    You’ll often see the phrase frontier model. It simply means one of the most capable AI models available at the moment.

    The frontier keeps moving. Today’s frontier model is next year’s ordinary one. The term shows up in three places:

    • Companies use it to describe their best work.
    • Industry groups use it too. OpenAI, Anthropic, Google, Meta, Microsoft, and Amazon belong to one called the Frontier Model Forum.
    • Governments use it in safety rules. The U.K.’s AI Security Institute tests frontier models before they’re released, and California passed a law in 2025 setting rules for companies that build them.

    Why there are so many new models

    New models arrive constantly because the labs are competing hard, and each new version is a chance to be faster, cheaper, or smarter than a rival. In 2026, each major lab has released a notable new model every few weeks to every couple of months.

    September 2026 alone brought:

    • Claude Fable 5.1 (Sept. 1)
    • Gemini 3.8 Flash (Sept. 2)
    • GPT-6 Astra (Sept. 3)
    • Grok 4.7 (Sept. 21)
    • GPT-6 Sol and Luna, and Claude Opus 5.5 (Sept. 22)

    Older models are retired over time, and apps usually switch to the newer ones automatically.

    Open and closed models

    One more term you may see: open-weight or open models. These are AI models a company lets anyone download and run on their own computers, often for free. DeepSeek, Qwen, Meta’s older Llama models, and Moonshot AI’s Kimi K3 are examples. Kimi K3, released in July 2026, is one of the largest AI models ever made freely available this way.

    The assistants most people use, including ChatGPT, Claude, Gemini, Grok, and Meta AI, run closed models. You can only use them through the company’s own app or service.

    What this means for you

    • You don’t need to memorize the names. When one shows up in the news, check the family (which company), the number (how new), and the tier (how big).
    • Free versions are real AI, just usually a smaller engine. For everyday questions, they’re often all you need.
    • Newer doesn’t always mean noticeably better for you. Many improvements matter most for complex work like writing software.
    • Company claims are claims. When a lab says its new model makes “half as many mistakes,” that’s the company’s own test. Keep checking anything important.
    • Try more than one. Asking the same question in two different assistants is a great way to see how they differ.

    Want to go further?

    • An opinionated guide to which AI to use (Ethan Mollick, One Useful Thing, July 2026). A respected Wharton professor’s regularly updated guide to the major AI tools. It’s more detailed and aimed at people already using AI for work, and it predates some of the newest models above.
    • AI glossary (TechCrunch). A reference for looking up individual AI terms. Some entries are technical.
    • Our Resources page, including an eight-minute video on how chatbots work.

    This guide was prepared with AI assistance, including Anthropic’s Claude, one of the models it describes. We’ve tried to describe every company the same way.