Smart and saved setups
In Smart mode, Smart picks a model for each message within two controls: Cost (Free, Pennies, A little, Some, Moderate, Generous, No limit) and Speed to Intelligence (Fastest to Strongest, with Balanced in the middle). New accounts start on Free and Balanced. Both are in the model menu and in Settings › Models › Smart and Manual. In Manual mode you set thinking, web search and styling yourself, and Smart still picks the model unless you choose one.
Mosaik fetches OpenRouter's model list once a day, and Refresh models fetches it at once. A model new to the list answers a short check before Smart picks it. Until it passes, Smart leaves it out, and you can still choose it yourself.
A saved setup is one you switch to from the model menu: Smart with a Cost and Speed to Intelligence of its own, or one model always. Make one in Settings › Models › Saved setups. Its window has a model for each job: chat, code, writing, maths, research, pictures, video, voice and the rest. Each job follows the setup's own controls until you give it a model of its own. Every change is saved at once.
A saved setup can also keep to its own privacy: skip providers that train on or store your chats, zero data retention only, or both.
Changing the model for one chat
The model button at the right end of the message box opens the model menu for the open chat: Smart's two controls, your saved setups, and Use one model, a search with your pinned, recent and all models. A model picked there is for this chat alone, until you change it. New chats start with Smart, or with the model picked on the New Chat page.
The model browser
All models, at the foot of the picker, opens OpenRouter's whole list of models. It has a tab for each kind of model (text, image, audio and more), filters for Free, Tool calling, Reasoning and Pinned, and sorts by newest, name, price or context size. Pick a model to see its context window, its prices, what it accepts, when it was added and what it can do: Vision, Audio, Reasoning, or No tools for models that can't use search or the exact tools. Pin the ones you use so they appear in the picker.
Reasoning
Some models can think before they answer. For those, a control next to the model button sets how much:
| Auto | The model decides how hard to think about each question. |
|---|---|
| Off, Low, Medium, High | A fixed effort. More thinking is slower and costs more tokens, and often gives better answers to hard questions. |
The choice is remembered for each model. When a model shares its reasoning, you can read it in the work line above the answer.
Providers
Most models on OpenRouter are served by several companies, at very different speeds. The same model can stream at twenty words a second from one and two hundred from another. Settings › Models › Advanced › Providers chooses how OpenRouter picks:
| Fastest | The default. The quickest provider, usually for a little more money. |
|---|---|
| Balanced | OpenRouter's own choice, which leans towards price. |
| Cheapest | The lowest price. |
The same section can let Mosaik retry with another free model when a free model gives no answer at all. It only switches before any text has arrived, so two models' words are never mixed in one answer.
Testing a model
Recommended in Settings › Models lists the families Mosaik is tuned for, by price. The Test button beside each runs a small task that needs a tool and checks the model called it and used the result. Worth doing before you rely on a new model for search or sums.