Choosing your model
Run the agent on Claude or GPT — and when it matters.
Databasin One isn't tied to a single AI model. The agent runs on whichever large language model you point it at — Claude or GPT — and you choose by picking an LLM connector and a model. Most of the time the default is fine; this article is for the times the choice matters.
How the choice works
Two settings decide which model runs the agent:
- The LLM connector. Databasin One talks to models through an AI & LLM connector. Pick one in the connector picker in the chat sidebar. An Anthropic connector runs the agent on Claude; an OpenAI connector runs it on GPT.
- The model. Within that connector, pick a specific model. Your choice is remembered per project, so each project can keep its own default.
If your project has no AI & LLM connector, the picker prompts you to create one before you can chat. A single connector is enough to get started.
The model picker
The picker offers a curated short-list of the latest Claude models, plus an escape hatch:
| Option | Good for |
|---|---|
| Claude Opus | The heaviest reasoning — complex analysis, tricky multi-step SQL. |
| Claude Sonnet | The balanced default — fast and strong for most questions. |
| Claude Haiku | The quickest and lightest — simple lookups and summaries. |
| Other… | Type any model ID by hand — including a GPT model when your connector is an OpenAI one. |
These track the current Opus, Sonnet, and Haiku releases. Sonnet is the default. If you've selected an OpenAI connector, use Other… to name the GPT model you want.
Claude and GPT work a little differently
Both families reach good answers; they just get there differently, and the agent adapts to each:
- Claude can batch tool calls — it may read the schema and a skill in the same turn, which tends to be a touch quicker.
- GPT works one tool at a time, so the agent gives it a higher iteration budget to finish the same job.
You don't have to manage any of this. It's worth knowing only so the two feel slightly different when you watch the agent work.
When the choice actually matters
For everyday questions, the default Sonnet is the right call. Reach for something else when one of these is true:
- You need more capability. Dense reasoning, gnarly SQL, or long documents benefit from a stronger model like Opus (or a top-tier GPT model).
- You want speed and lower cost. Routine lookups and short summaries run fine — and cheaper — on a lighter model like Haiku.
Whichever you choose, agent usage draws on your credits, and heavier models cost more per turn. If you're running a lot of background work or large analyses, the model choice shows up on your bill.