Choosing models
Found this helpful? Share it:
Found this helpful? Share it:
Every model has different strengths: some are fast, some dig deep, and some shine at specific work like coding or creative writing. idapt gives you more than 200 to choose from, each labeled with a cost band, so you can match the model to the task 🧠
Open the model selector in the composer toolbar, below the message box, and pick any model. Your next message uses it, and every earlier message stays exactly as it was answered.
This is handy when you start a chat on a fast model and then want a more capable one for a tricky follow-up. Switch back any time.
Every model carries a cost band from $ to $$$$, based on how much it costs to run per message. Lower bands are cheaper and usually faster, higher bands are pricier and usually more capable.
On a new account the $ (Affordable) models are the ones your welcome credit covers. Pricier models Usage Creditsunlock with your first top-up: until then the picker shows an Add credits prompt, and one top-up unlocks every model permanently. See Plans, credits, and the welcome credit.
Auto is the default. It picks a concrete model for you based on the speed dial next to the selector: a cheaper, faster model when the dial is on Low, the paid mid-tier model on High, and the frontier flagship on Max. It resolves from the dial, not from reading your message, so the choice is predictable.
Auto is the best choice for most conversations. You get strong answers without thinking about model names at all.
Override Auto any time by selecting a specific model in the composer. Switch back to Auto whenever you like.
Next to the model selector is a small dial labeled Low, Medium, High, or Max. It is the single control for how hard the AI works before answering, and it also shapes how the assistant talks back.
Low: quick, concise answers. Best for short questions, drafts, and back-and-forth. The assistant keeps replies short and skips unnecessary background.
Medium: balanced depth and speed. The default for most conversations.
High: thorough, detailed analysis. Ideal for tricky problems, code reviews, and long planning. The assistant takes more time, explores edge cases, and surfaces trade-offs.
Max: the strongest model working at full depth. For the hardest problems where you want every constraint checked and every alternative considered.
When Auto is selected, the dial chooses the model (cheaper on Low, more capable on Max). When you have chosen a specific model, the dial controls how much that model reasons and the behavior it adopts, so even a fast non-reasoning model answers differently on Low versus Max.
Flip the dial mid-chat and the next response picks up the new mode. The assistant is told you switched to Low, Medium, High, or Max so it adjusts tone and depth right away.
Open the Models directory from the sidebar More menu, from the More models button inside the model picker, or by pressing Cmd+Kand searching for "Models".
Filter by author (Anthropic, Google, OpenAI, xAI, and more), cost band, speed, a context-length slider, capabilities (vision, audio, reasoning), and status. Sort the results, including by coding strength, to line up candidates for a task. Each card shows the context length, pricing, and supported features at a glance.
Click any model card to open its full details: benchmarks, a pricing breakdown, and a button to start a new chat on that model.
If you pair a computer running local models, those models show up in the picker as a free provider: prompts and responses never leave your machine, and there is no per-message cost. Models you already pulled with Ollama appear too, with no re-download. The directory has a filter to show only the models that fit your hardware. See Computers.
Context size is how much of the conversation a model can see at once. A model with a 128K context can reference far more history than one capped at 8K, and some models reach up to a million tokens.
Larger contexts let the AI hold on to earlier parts of a long chat, but they also cost more per message. For long conversations, turn on Auto-Compact so idapt summarizes older messages before you hit the limit. See Chat settings and toggles.
Related articles
Retrying a response
Regenerate a reply, optionally with a different model.
How we score models
Understand the capability score, the named public benchmarks, and the speed, reliability, and cost numbers on every model page.
Chat settings and toggles
Customize each conversation from the composer: tool toggles, autonomy, compaction, and a spend limit.
Was this helpful?