Change one line. Get proof, not just an answer.
Run the exact same extraction call against OpenAI and Interfaze. Interfaze hands back a confidence score and the spot on the page it read each value from. OpenAI hands back an answer you have to trust.
const client = new OpenAI({ - baseURL: "https://api.openai.com/v1",- apiKey: process.env.OPENAI_API_KEY,+ baseURL: "https://api.interfaze.ai/v1",+ apiKey: process.env.INTERFAZE_API_KEY, }); - model: "gpt-5.6-terra"+ model: "interfaze-beta"
Pick a document, or upload your own
How this comparison works — same prompt, same schema, same code path, and where Interfaze actually loses. Read the fine print ↓
Both panels come from one function, callProvider, issuing the same OpenAI-SDK request. Only three things differ: the base URL, the API key, and the model string. Interfaze implements the OpenAI chat-completions spec, which is what makes that possible — the migration really is a config change.
Both providers get the identical prompt (“Extract the requested fields from this document. Return only values present in the document. If a field is not present, return null for it.”), the identical JSON schema, and temperature=0.
On confidence and boxes. These are not from a special endpoint or parameter. Interfaze returns a precontextarray on the ordinary response holding its OCR encoder's raw per-word text, bounds and confidence. A general model has no equivalent, so its panel is empty — not because we withheld anything, but because there is nothing to show. We could have asked both models to emit a confidence field in the schema; both would have complied and one would have been guessing, which would prove nothing.
precontext carries no pointer back to your schema fields, so the badge-to-field mapping is a string match we perform against the OCR text. Fields with no verbatim match get no badge rather than a guess.
On cost — read this before quoting a number. Cost is computed from the token usage each provider actually reports, at rates read from their public pricing pages on 2026-07-16: OpenAI $2.5/$15 and Interfaze $1.5/$3.5 per 1M input/output tokens. We make noclaim that Interfaze is cheaper per call — it usually isn't. It runs full OCR internally, so it consumes several times more input tokens than a general model, and a lower per-token rate does not offset that. Interfaze's case is accuracy and verifiable output, not price; the per-panel cost is shown so you can see the real trade rather than a spun one.
Single documents, single runs. This is a demonstration, not a benchmark — one run proves a shape, not a win rate.