01
Identity confidence
Behavioral consistency against a controlled reference
Independent endpoint verification
Can you trust the AI model behind your API?
A model name is a claim, not a quality guarantee. ModelTrust measures capability integrity and reports calibrated evidence about the endpoint you receive.
VERIFICATION / MT-1042 COMPLETE
Trust score
82/100
B+
Behavior appears broadly consistent with the claimed model. Tool-call reliability requires further sampling.
Identity confidence88High
Capability retention81Stable
Tool calling74Review
Context integrity85Stable
Illustrative report · Not a live verification
Why now
Cheap access creates expensive uncertainty.
02
Capability retention
Reasoning and instruction-following under repeatable tests
03
Interface integrity
Tool calls, context handling, and token accounting
04
Operational quality
Endpoint latency, availability, and response reliability
Project status / v0.2.0
Building in public, with evidence.
Model Verification MVP is 35% complete. Follow shipped work, verification evidence, and the next development target.
Principle 01
Pay to test. Never pay to rank.
ModelTrust is designed to remain structurally independent from the providers it evaluates. Findings should become transparent, reproducible, and resistant to commercial influence.