Behavioral model verification

Test what an AI endpoint actually behaves like

Run a single-request behavioral fingerprint against an external AI API. Model Gate creates the challenge, your browser calls the external endpoint directly, and a fixed evaluator scores the response.

Verify an endpoint How it works

One external requestExternal key stays in your browserVersioned evaluator

● ● ●

External endpoint
      ↓ browser AJAX
one transparent generated challenge
      ↓
behavioral fingerprint
      ↓
Model Gate judge
      ↓
score + identity signals

Methodology

A behavioral screen, not a self-identification quiz

The test combines factual recall across time, self-report consistency, abstention behavior, confidence calibration and strict structured-output compliance. The model name returned by an API is recorded as metadata, not treated as proof.

Challenge

A fixed administrator-selected evaluator creates one compact challenge with a hidden answer key.

Browser request

Your browser sends the challenge directly to the external HTTPS endpoint. The external API key is never submitted to Model Gate.

Scoring

Model Gate evaluates the returned answer and calculates recall, recency, hallucination resistance, calibration and instruction-following scores.

Behavioral verification cannot cryptographically prove the underlying model identity. Public browser tests can be modified by the tester; community verification therefore requires consistent published results from multiple independent Model Gate accounts.

Community verified

Endpoints with consistent independent results

An endpoint appears here only after at least three different Model Gate accounts publish tests within 30 days, with an average score of at least 80 and low score variance.

Endpoint	Claimed model	Protocol	Evaluator	Average score	Independent testers	Last tested
No endpoint has enough independent published tests yet.

Latest published tests

Recent behavioral assessments

Run your own test →

CONSISTENT

`deepseek-v4-flash`

https://api.model-gate.com/v1/chat/completions

Recall: 83
Recency: 82
Hallucination resistance: 83

gpt-5.6-sol · model-verification-single-request-v1.2 · Sep 6, 2026 16:39

100

PLAUSIBLE

`deepseek-v4-flash`

https://api.model-gate.com/v1/chat/completions

Recall: 100
Recency: 100
Hallucination resistance: 100

gpt-5.6-sol · model-verification-single-request-v1.1 · Sep 6, 2026 09:53

Verify an AI endpoint with one request

Choose your Model Gate billing key, enter the external endpoint and run a browser-side test.

Open Model Verification