# OpenAI

> Connect OpenAI's own API with a key, and what the Hub does with its models.

OpenAI's GPT models through your own OpenAI account, billed per token to that account.

## Set it up

1. Make an API key at https://platform.openai.com/api-keys.
2. In **Settings → Model Hub → Providers**, select the **OpenAI** tile and paste the key into **API key**.
3. Select **Test**. The dialog says *Connected* and how many models were found.

## Models and prices

The Hub carries a short table of current models with their listed prices, and shows the ones your account can see:

| Model | In, per million | Out, per million | For |
|---|---|---|---|
| `gpt-6.1-sol`, `gpt-6-sol` | $2.00 | $10.00 | Most work |
| `gpt-6-luna` | $0.10 | $0.50 | Bulk and quick work |
| `gpt-6-astra` | $10.00 | $50.00 | The hardest problems |
| `gpt-5.4-mini` | $0.75 | $4.50 | |
| `gpt-5.4-nano` | $0.20 | $1.25 | |
| `text-embedding-3-small`, `text-embedding-3-large` | $0.02, $0.13 | | Embeddings |

Prices were checked on 2026-10-06. OpenAI reports no cost per call, so the ledger's cost is the listed price and **Billed** stays empty. A model outside the table can still be named in a tier, but it has no price, so a call on it is refused as unpriced until the table catches up.

## Reasoning

The **Reasoning** option on a model's row is sent as OpenAI's `reasoning_effort`: *Low*, *Medium* or *High* as they are, *Off* as `none`, and *Model default* as nothing at all. See [Set up tiers and limits](../tiers-and-limits.md).

## Web research

Research roles use OpenAI's own web search on the Responses API. Sources come back with the reply.

## Related

- [Add a provider](../add-a-provider.md)
- [Privacy and costs](../privacy-and-costs.md)