Should an AI assistant tell you which model it is?
Yes. An honest assistant names the model writing this reply, the one doing heavy work such as builds, and why those two can differ. A brand story or a scold that “that is not the question” is a dodge. You need the names because the company behind a model sets what happens to what you send.
Why does “which model are you?” deserve a straight answer?
The question is not small talk. The model that writes a chat line may not be the one that writes documents, checks a build, or does hard thinking. A quicker model can answer when the usual one is busy. Each vendor’s terms, retention, and training policy apply to the traffic that vendor sees.
If the assistant will not name those roles, you cannot choose a tool that fits the job. A simpler local chat app with one fixed open-source model is the better choice when you need a single, inspectable stack and do not want a router at all.
What should an honest answer actually name?
A useful reply names three things. First, the model composing this sentence. Second, the stronger model used for builds, documents, and hard thinking, if that is a different engine. Third, the reason they differ: cost, speed, or a setting you chose.
It should also say when pictures, video, or live voice always stay on the host’s own engines, and when your own provider key is billed to you with no markup. Ask again after you change the brain setting; the names should change with it. GROX, if asked which model it is, names the model writing that reply and the one doing builds.
- This reply
- The engine that composed the words you are reading now, including jokes, refusals, and short answers.
- Heavy work
- The stronger engine used for builds, long documents, and hard thinking when that work is split from chat.
- Why they differ
- A quicker model for chat, a stronger one for builds, a backup when the usual one is busy, or a brain you picked in settings.
- Your key
- When you bring your own provider key, that provider bills you; some options run only on your own key.
Which dodges should you distrust?
Distrust a brand story in place of a name. “I am your copilot” or “I am a helpful assistant” is not a model. Distrust “that is not the question.” Distrust a list of every model the company has ever used, with no pointer to this turn. Distrust silence after you change the brain setting.
A local single-model tool will always give the same name. That is a real trade-off: less routing, less capability, and a clearer paper trail. Use that when the job is private notes, not deployed apps.
| You asked | Straight answer | Dodge to distrust |
|---|---|---|
| Who wrote this reply? | A named model for this turn | A personality or product slogan |
| Who does the heavy work? | The build or thinking engine, named | “We use several partners” |
| Why do they differ? | Speed, cost, or your setting | “That is not the question” |
| What if I switch brains? | New names that match the setting | The same vague line as before |
When is a simpler tool the better choice?
If you only need one model, one vendor, and files that never leave your machine, a desktop client with a local open-source model is simpler. You already know the name; you do not need a router. You also lose connectors, a cloud computer that signs in step by step, and a build tab that deploys.
If you do want routing, pick a product that shows the credit price of each brain before you choose, lets you ask which model is writing, and keeps pictures, video, and live voice on stated engines. GROX puts those brains in Settings → Brain, with its own engine as the default, or Claude, GPT, DeepSeek, Kimi, MiniMax, GLM, Qwen, or your own open-source model.
Common questions
Should an AI assistant tell you which model it is?
Yes. It should name the model writing this reply, the model doing builds or hard thinking if that is different, and why they differ. A slogan, a refusal, or a history of every partner is not an answer. You need the names because the company behind the model decides what happens to what you send.
Why can the chat model differ from the build model?
Chat often uses a quicker engine so short replies stay cheap and fast. Builds, documents, and hard thinking often use a stronger engine. A backup can answer when the usual one is busy. An honest product says so, shows a credit price before you switch, and names both engines when you ask.
What happens to what I send if I pick another company’s model?
The company behind that model sets retention, training, and logging for traffic it sees. If you use your own provider key, that provider bills you with no markup, and some options run only on your own key. Host pictures, video, and live voice may still run on the host’s own engines even when chat uses another brain.
When should I skip a routed assistant altogether?
Skip it when one local model is enough and you want every token on hardware you control. A single-model desktop app is then the better tool. Choose a routed assistant only if you need named engines for chat versus builds, optional third-party brains, and a clear answer when you ask which model wrote this line.