Models

verified.create, verified.eval, and verified.extract work with any text model your gateway key can call — verification is a layer on top of the maxmodel.com gateway, not a separate model. You pass a model id; the gateway routes the EXTRACT step to that provider, then the deterministic check runs the same way regardless of which model you picked.

Which models are eligible

Any model the gateway exposes on its chat or responses endpoints (i.e. text generation). As of the latest gateway snapshot that’s 63 models across 8 providers:

The live list is the source of truth. The full, current set is published as JSON at maxmodel.com/models/list.json — fetch it at build time if you want a dropdown that never goes stale. Models on the image, video, embedding, tts, and transcription endpoints are not eligible for verified output (there’s no text answer to ground).

Picking a model

The verification result is identical no matter which model you choose — the check is deterministic. So pick on the usual axes (quality, latency, cost), with two MaxModel-specific notes:

• For large verified.eval runs, use a fast model. A dataset run is one model call per case; a reasoning/responses-endpoint model (e.g. gpt-5.5-pro) makes a 200-case batch slow. A fast chat model like claude-haiku-4-5-20251001, gpt-5-mini, or gemini-2.5-flash keeps eval runs quick and cheap.
• responses-endpoint models are higher-latency. gpt-5.5-pro, gpt-5.3-codex, and gpt-5.1-codex-max route through the Responses API and can take 10–20s for a single verified call. Great for a final answer; not what you want in a tight loop.

A model that quotes its sources cleanly will ground more of its own output — but that’s a quality difference in the answer, never in the check.

Notes

• The model you pass must be callable by your gateway key. An unknown or unentitled id comes back as a gateway error (see Errors).
• verified.create is non-streaming regardless of model — verification needs the full claim set. Use chat.completions.create for token streaming without verification.