How to Tell Which AI Model Version You Are Using (and Why It Sometimes Changes)

You can usually find which AI model you are using in under a minute, because ChatGPT, Gemini and Claude each put the name in a picker or menu, even if automatic switching sometimes changes it.

In ChatGPT, look for the picker at the top of the chat or inside the composer

OpenAI places the picker at the top on phones and in the message box on web.

OpenAI’s release notes describe a simplified picker with options labelled Instant, Medium, High, and, for Pro plans, Extra High, Pro Standard and Pro Extended. On iOS and Android it sits at the top of the conversation; on web, in the message composer. The level shown there is the clearest statement of what will answer you.

Read those labels for what they are: a balance of speed and reasoning effort, not a full technical model name. OpenAI renamed its older Thinking options into this simpler set. Also check Settings: users can decide whether Instant auto-switches to Medium for higher reasoning when required, under General in Settings. If answers feel inconsistent, that toggle is the first place we would look.

In Gemini, the current model sits at the bottom of the chat box

With a Google AI plan you can tap that name and switch, and the choice lasts for the conversation.

Google’s Gemini Apps Help page says that with a Google AI plan, you can manually switch models: on a computer, go to Gemini, click the current model at the bottom of the chat box, then click the model you want. Locked models need an eligible plan upgrade. You can also pick a thinking level of Low, Medium or High, defaulting to Low.

Plan matters more here. Google’s table says accounts with no Google AI plan get Auto or Flash-Lite in chats, Google AI Plus adds Flash and Pro, and higher plans add stronger models such as Argon. Gemini keeps using your chosen model and level for the rest of that conversation, even if you return later. Our pillar guide on how AI model updates work across ChatGPT, Gemini and Claude explains the families behind those names.

In Claude, open the model menu next to the send button

That one menu controls the model, the effort level and thinking together.

Anthropic’s Help Center says the model menu next to the send button controls three things: which Claude model you are chatting with, how much effort it puts into each response, and whether it uses thinking. Clicking the current model name opens that menu, so the name beside the send button is your at-a-glance answer.

Two warnings are worth knowing. If a model or effort level is missing on an Enterprise plan, your administrator may have turned it off, and an administrator can also set the model every new chat starts on. For everyone else, glance at the menu before an important conversation, not after. Our comparison of what ChatGPT’s free and paid tiers actually give you makes the same point: the picker is part of what your plan buys.

The three pickers side by side

One table covers where each service shows the model and who gets to choose.

If the vocabulary feels confusing, keep our plain-English AI terms glossary open while you explore. It makes the labels far less intimidating.

Automatic switching is documented, and it explains most mystery changes

Each provider describes situations where the app, not you, decides which model answers.

Here is the part most people never get told. In Gemini, auto model selection routes prompts to the most efficient model and thinking level, and Google says most prompts will use Flash-Lite while harder ones may use Flash. Reach your usage limit on a paid plan, and your chat will continue with Flash-Lite. In Claude, automatic switching is active by default for Fable models: if a safety check blocks a request, it can be re-run on an Opus model in the same conversation, you see a notice, the answer is labelled with the model that produced it, and the picker stays on Opus until you switch back.

ChatGPT’s version is the Instant toggle: Instant can auto-switch to Medium when more reasoning is required, if you leave it on. None of this is a glitch. It is three companies making the same trade, saving their strongest models for prompts that seem to need them and using faster ones the rest of the time.

A different answer to the same prompt usually means a different model, level or moment

Before blaming your wording, check the picker, the level, and whether a switch intervened.

So why did yesterday’s perfect prompt produce a flatter answer today? Work through the boring explanations first. The picker may sit on a faster option from an earlier conversation. A thinking or effort level may differ, and Anthropic says higher effort means more thorough responses that also use more tokens. You may have hit a usage limit that moved you to a lighter model, as Google documents. Or a safety fallback may have moved your Claude conversation from Fable to Opus, with only a small notice to mark it.

The honest answer is that the same prompt was never a guarantee of the same answer. Checking the visible name first turns a vague suspicion that the AI got worse into a specific, fixable fact: wrong picker, wrong level, or a switch you can reverse.

What this means

Ten seconds checking the model name saves more frustration than any prompt trick.

There is a particular irritation that only AI users know: you finally craft a prompt that works beautifully, return to it a week later full of confidence, and the magic has gone. The answer is shorter, or blander, or oddly confident about the wrong thing. Your first instinct is to blame yourself, then the prompt, then the whole technology. Meanwhile the culprit is sitting quietly at the top of the screen, or the bottom, or beside the send button, wearing a slightly different name than last time.

That is why, in our view, model-checking deserves to become a reflex like checking which account you are emailing from. It costs almost nothing. Before important work, glance at the picker and confirm you are on the model and level you meant to use. After a strange answer, check whether a limit, a toggle or a safety fallback moved you. If you share prompts, tell people which model and level the prompt was tuned on, because a recipe that names its oven temperature travels better than one that does not.

We would also defend the providers a little. Automatic routing is not purely a cost-cutting trick. Most everyday prompts genuinely do not need the deepest model, and routing them to a faster one gives a quicker answer and preserves your usage for prompts that do. The deal only turns sour when the switch is invisible. Encouragingly, all three help centres now document where to look, and Claude promises a notice and a labelled answer when a fallback switch happens. Use those tells, keep a tiny note of the model and level beside any prompt you truly depend on, and your AI sessions get calmer, more predictable and far easier to troubleshoot on the days when something quietly changes underneath you again.

Bottom line

Check the picker before the prompt, and most mysteries solve themselves.

ChatGPT puts its picker at the top of mobile chats and in the web composer, Gemini puts the model at the bottom of the chat box, and Claude puts its menu by the send button. Know your plan’s limits, watch for automatic switches, and next time an answer changes character, you will know where to look.

Related Guides

The pillar, plan comparison and glossary that complete this cluster.

• AI Model Updates Explained: How ChatGPT, Gemini and Claude Get New Versions
• ChatGPT Free vs Plus: What You Actually Get
• AI Terms Glossary: Tokens, Prompts, RAG and Hallucinations

Sources

The official help pages behind every picker location above.

• ChatGPT release notes — OpenAI Help Center
• About Gemini models — Gemini Apps Help
• Change the model, effort, and thinking settings — Claude Help Center
• Why Claude switched models in your conversation with Fable models — Claude Help Center