Superion S1 answers with a probability.
Ask it a question about any text. It tells you how likely each answer is, in one fast call, so your software can act on it.
Is this review positive?
“Arrived two days late, but it works perfectly and my daughter loves it.”
Illustrative output.
Highest average score on S1MB.
S1MB is an open benchmark built for models like this one: 137 English tasks covering fact checking, safety, reasoning, relevance and grading. The maintainer validates every submission. Superion S1 scored 74.36, ahead of the next model by 2.2 points, with all 26,269 decisions validated.
Task Average on S1MB English v1, from the public S1MB results dataset, snapshot of 2 November 2026.
Three ways to ask.
Is it true?
Check facts, flag unsafe content, enforce a policy.
Does this reply give medical advice?
Which one?
Route requests, classify intent, pick the right answer.
Which team should get this ticket?
How much?
Grade answers, rank results, rate severity on your own scale.
How helpful is this answer, 1 to 5?
Illustrative outputs.
The Superion S1 family.
One fast decision model, tuned for the places where decisions matter most. Each member runs as its own model, with its own measured results.
Available soon on OpenRouter
Superion S1
The general model. Fact checking, safety, routing, grading and relevance, in English. Highest Task Average of 213 models on S1MB.
In development
Superion S1 Legal
A variant being tuned for US contract clauses. Results will be published on public test sets.
In development
Superion S1 Neuro
For physical AI. The decisions a robot makes every second: who is speaking to me, what just changed, is the task done. Designed to work inside ELIOT Robot OS.
Need a version for your industry? We tune Superion S1 to your data.
Try it free for 24 hours.
Ask for a key below and we email it to you right away. It works with every request type for 24 hours. No card, no account.
Your trial key
The same model and endpoint we run in production, with a smaller quota.
- Valid for24 hours
- Requests per minute30
- Keys per email1 per week
- PriceFree
Ready for production? Two ways to run it.
API on OpenRouter
Sign up and pay through OpenRouter, the same way you use any other model there.
Output tokens are free.
Available soon on OpenRouterLicense it, tuned to your industry
A general model gives general results. On request, we build a custom decision model for your industry: your categories, your rules and your own data. We measure it on your cases before you sign. Then it runs on your infrastructure or inside your product.
Insurance claims, support routing, content moderation, compliance review, retail catalogs, robotics. If your software has to decide, we can tune it.
Email us about licensingalvaro.lopez@punkytigerlabs.com