Model names decoded

GPT-6.1 Sol. Claude Opus 5.5. Gemini 4 Argon. Sounds like galaxies, but there’s a logic behind it that’s the same at almost every maker.

After this lesson you can: break down any model name, tell the big models from the small ones, and find out which model is running in your app.

App ≠ model

ChatGPT, Claude, Gemini, Grok and Mistral Vibe (formerly Le Chat) are apps. Which model runs inside depends on your plan and settings. A free user may be using a very different model from a paying one, even though they both say “ChatGPT”.

Extra confusing: “Claude” and “Gemini” are both the app and the model family.

How to read a model name

Almost every name is made of the same building blocks:

flowchart LR
  A["Family<br/>GPT · Claude · Gemini"] --> B["Version<br/>6.1 · 5.5 · 4"] --> C["Size<br/>Sol · Opus · Argon · Flash"]

1. Family: who makes it. GPT (OpenAI), Claude (Anthropic), Gemini (Google), Grok (SpaceX Super Intelligence, formerly xAI), Muse (Meta), Mistral, DeepSeek, Qwen (Alibaba).
2. Version: higher is newer, but only within the same family. “Claude Opus 5.5” isn’t worse than “GPT-6 Astra” because the number is lower.
3. Size: large (smartest, slower, more expensive) or small (fast, cheap). The most important word for you.

The size dictionary

As of 1 November 2026:

Maker	Large			Small
OpenAI	Astra	Sol	Terra	Luna
Anthropic	Fable	Opus	Sonnet	Haiku
Google	Argon (Carbon still limited)	Pro¹	Flash	Flash-Lite
Meta	Wonder	Spark		Glimmer
Mistral	Large	Medium	Small	Ministral
DeepSeek	Pro			Flash
Qwen	Max	Plus		Flash

¹ Google’s Pro model is still from the previous generation (Gemini 3.1 Pro). Not to be confused with the “Google AI Pro” plan.

Pro tip

Bigger isn’t always better. New mid-sized models often beat older top models. Anthropic itself recommends starting with Opus 5.5 for most tasks, not the larger Fable. And OpenAI describes GPT-6.1 Sol as offering “near-Astra performance for complex work at a lower cost”.

Five pitfalls

1. Each size gets its own updates. At OpenAI, GPT-6 Astra and GPT-6 Luna sit alongside GPT-6.1 Sol and GPT-6.1 Terra; at Google, Gemini 4 Argon sits alongside Gemini 3.8 Flash.

2. Version numbers are labels, not numbers. With Grok, Grok 4.20 came out in March 2026, before Grok 4.3 and 4.5. So don’t read “4.20” as “4.2”.

3. Thinking is increasingly a setting, not part of the name. There used to be separate “thinking” models (such as OpenAI’s o3 or DeepSeek R1). Now it’s usually a switch: effort from low to maximum, or “adaptive thinking”. In ChatGPT, for example, you choose a faster or smarter mode.

4. Nicknames aren’t models. Nano Banana is Google’s image model, not the same as Gemini Nano. And according to Anthropic, Claude Mythos 5.1 is the same model as Fable 5.1, just with different safety settings and only available to selected organisations.

5. Open or closed. Some models you can download and run yourself (open weights), such as OpenAI’s gpt-oss, Meta’s Muse Spark Open, Mistral Large 4 and the DeepSeek V4 models. Closed models, such as GPT-6 Astra, Claude Fable and Gemini 4 Argon, you can only use via the app or API. Open weights doesn’t mean “anything goes”: every model has its own licence.

Building something? Pin your model

Names like -latest or a bare family name automatically move to the newest version: one update and your prompt behaves differently. In production, use a fixed version (a snapshot, for example claude-opus-5-5 or an ID with a date) and only switch after your own test.

Which model comes with your plan?

According to the makers themselves (as of 1 November 2026; changes often):

Claude

Plan	Haiku	Sonnet	Opus	Fable
Free	✓	✓	–	–
Pro	✓	✓	✓	only with usage credits
Max and Ultra	✓	✓	✓	up to half of your weekly limit

Gemini

Plan	Model
Google AI Plus	the Flash model
Google AI Pro and Ultra	Gemini 4 Argon (Ultra with the most generous limits)

ChatGPT: free users get a smaller model than paying users. OpenAI publishes the exact breakdown on its help pages and adjusts it regularly; check the app to see which model you get.

Pro tip

Which model are you using right now? In ChatGPT, Claude and Gemini there’s a drop-down menu at the top or bottom of your chat window with the model name or a mode such as “faster/smarter”. Click it to see your plan’s options.

Security

• Fake names are bait. “Free GPT-7” in apps, extensions and ads is a well-known malware trick; fake ChatGPT and DeepSeek apps that stole data have already appeared. Only use the official app and check the publisher in the store.
• The app decides where your data goes, not the model. In 2026, millions of conversations were left exposed at wrapper apps and aggregators (Chat & Ask AI, 302.ai).
• Building with a pinned model ID? Keep your API key in an environment variable, not in your code (managing secrets).

Checklist

0/4

• I know that the app and the model are two different things
• I can point out the family, version and size in a model name
• I know that bigger isn't always better
• I've looked up which model I use in my own app

Your progress is only stored in this browser.

Test yourself

1. 1What is Anthropic's smallest and fastest size?

   OpusSonnetHaiku

2. 2Is Claude Opus 5.5 worse than GPT-6 Astra because 5.5 is lower than 6?

   Yes, a higher number is betterNo, you can only compare version numbers within the same familyYes, but only for code

3. 3What does "open weights" mean?

   You can download the model and run it yourself (under its own licence)The model is free to use in the appThe model has no safety rules

Next lesson: which assistant should you choose, with figures instead of opinions.