# Frames F1

> Frames' research model, and when to choose it over MCP.

Frames F1 is our research model. Give it a task and a budget, and it does the whole job: it plans the research, buys data from several sources, checks what it bought and returns a finished answer with sources, receipts and a confidence score.

## What F1 does on each run

1. Plans the task and picks sources from the catalog.
2. Buys the data it needs, never going over the run's budget.
3. Checks each paid response and takes another pass when results are thin.
4. Writes the answer, as text or as JSON in a shape you define.
5. Returns receipts, a cost breakdown and a confidence score.

## MCP or F1?

| Choose MCP when | Choose F1 when |
| --- | --- |
| You work in Claude, ChatGPT, Cursor or another AI app | You're building an app, a pipeline or a scheduled job |
| Your own AI can do the reasoning | You want Frames to do the reasoning |
| You want the lowest cost | You want a finished, checked answer from one request |

## How to reach F1

- **From code:** `model: "frames-f1"` on the OpenAI-compatible endpoints, or `POST /v1/runs`. See the [API](https://frames.ag/docs/api.md) page.
- **From an MCP app:** the `frames_run_capability` tool. See [MCP tools](https://frames.ag/docs/mcp-tools.md).

## Cost

F1 is the premium path through Frames, and its price reflects what it does. Over MCP your own AI does the thinking and Frames bills only the data that arrived, plus 15%. With F1 the model does the thinking too: it plans the run, chooses the sources, reads every paid response and writes the answer, so its model work is billed on top of the data, with the same 15% on both. In return you get a finished, checked answer from one request instead of raw responses to assemble yourself.

A run never costs more than its budget: Frames sets the budget aside before the run starts and returns whatever was not used. A run that fails costs nothing, a completed run that bought no data and ended without a confident answer costs nothing, and paid calls that fail the delivery check are never charged. Every run reports its own split in `usage` (`model_cost_usd`, `tool_cost_usd`, `frames_fee_usd`, `total_cost_usd`), and the chat shows the same split under each answer. [Credits, pricing and receipts](https://frames.ag/docs/pricing.md) has the full breakdown.