VetEval & viggoVet
The Veterinary AI benchmark.
VetEval is an evaluation service that measures how well AI models perform on real veterinary medicine: expert-authored examination items across species and specialties, scores with confidence intervals, and a safety gate that caps any model producing advice a panel of licensed veterinarians confirms as harmful. The full methodology, scoring math, and leaderboard are published at veteval.ai.
viggoVet operates it. Here is why, and here is how we keep that honest.
Why a veterinary AI company built a benchmark.
When we walked through the clinic asking what an AI should do in each room, a prior question kept surfacing: is any of this reliable enough to be in the room at all? We were not alone in asking. In 2025, the American College of Veterinary Radiology and the European College of Veterinary Diagnostic Imaging formally called for an outside organization to evaluate and set guidelines for veterinary AI. No such organization existed. So we built the instrument, and then we built a wall between it and ourselves.
The wall, in five commitments.
- viggoVet models are never ranked on the VetEval public leaderboard. Not labeled, not listed, never.
- The item bank, gold answers, and transcripts stay with the benchmark team. They are never shared with viggoVet product teams and never used to train or tune anything viggoVet builds.
- No vendor funds VetEval’s operations or influences any score. Evaluation fees are uniform and published; publication is free and decided blind to the score.
- Being a viggoVet customer, including customers of our APIs, has zero bearing on any VetEval evaluation, positive or negative. Endpoints that incorporate viggoVet services are not eligible for the public leaderboard at all.
- VetEval’s scoring parameters and safety severity tiers are governed by licensed veterinarians and versioned with the dataset, so every published number can be traced to exactly what produced it.
These are not promises to take on trust; they are controls you can verify. The separation architecture, run conditions, and dataset governance are documented at veteval.ai/methodology.
What this means if you're a viggoVet customer.
You will never see us quote a VetEval score, embed the leaderboard, or wear a benchmark badge. That absence is the policy working. Our own products are validated separately and published as viggoVet research as that work completes, because a leaderboard measures what a model knows, and it cannot measure whether a monitoring alarm fires in time. If you are evaluating AI for your practice, from us or anyone else, use VetEval the way it is meant to be used: read the methodology, then look up the model behind the product you are considering.
Common questions
Why build a benchmark you refuse to appear on?
Because the profession needed the instrument more than we needed the placement, and an instrument its operator competes on is not an instrument. The absence is the price of the credibility, and we pay it gladly.
How do I know the firewall is real?
Verify rather than trust: run conditions are enforced by the evaluation harness itself, every published score pins its items, harness version, and grader panel, and the dataset governance is versioned and hashed. The controls are documented at veteval.ai under "How the teams are separated."
Does buying viggoVet APIs help a model score better?
No, and the rule goes further: an endpoint that incorporates viggoVet services is ineligible for the public leaderboard entirely, so the question can never even arise on the board.
Where are viggoVet's own results, then?
Published separately as viggoVet research as that work completes, with methods shown, and never as leaderboard entries.
New benchmarks, the day they land.
Get the new VetEval rankings when they publish: which AI models are ready for veterinary medicine, and which are not.
VetEval is operated by viggoVet and published at veteval.ai. For what viggoVet itself builds, start with the clinical AI tools.