Decisions, not text.
Superion S1 answers the small questions inside your product — yes or no, which option, how much — with calibrated probabilities, in one call, in well under a second.
#1 of 214 on S1MB English v1
Task Average 75.06 · Borda 89.82 — public S1MB leaderboard, 12 November 2026. See the leaderboard.
HOW IT WORKS
Send a text
A ticket, a comment, a lead, an agent's next action. Up to 32K tokens.
Ask typed questions
Yes/no, a choice among up to 255 options, or a score on 2–16 levels. As many as you need, in the same call.
Act on the numbers
One calibrated probability per question. Put a threshold on it and let your code do the rest.
One call, three decisions
A real request and its real response, from the docs.
REQUEST
{"model": "superion-s1-latest",
"state": "Message: The laptop arrived broken. I want my money back now, this is unacceptable.",
"questions": {
"refund": {"type": "noul",
"instructions": "Does the customer ask for a refund or a replacement?",
"criteria": {"true": "yes", "false": "no"}},
"sentiment": {"type": "choice",
"instructions": "Overall sentiment of the message.",
"criteria": {"pos": "positive", "neg": "negative", "neu": "neutral"}},
"urgency": {"type": "score",
"instructions": "How urgent is a reply?",
"criteria": ["not urgent", "somewhat urgent", "very urgent"]}}}RESPONSE
{"answers": {
"refund": {"type": "noul", "noul": 0.9998},
"sentiment": {"type": "choice", "choice": "neg",
"probabilities": {"pos": 0.0003, "neg": 0.9941, "neu": 0.0056}},
"urgency": {"type": "score", "score": 1.9393,
"probabilities": {"0": 0.0, "1": 0.0607, "2": 0.9393}}},
"usage": {"input_tokens": 135, "output_tokens": 0}}
Abridged from the real response of 13 Nov 2026 (model superion-s1-v6-int4e-vllm-kv1@sha256:e6dc49dc). Full request and response.
What teams build with it
Agent guardrails
Check every action before your agent runs it.
ASKS
Does it match the request? Is it irreversible? How risky is it?
COST*
≈ 44 USD per million actions
Ticket triage
Queue, urgency and refund intent, in one call, before a human opens the ticket.
ASKS
Which queue? How urgent? Is money at stake?
COST*
≈ 41 USD per million tickets
Content moderation
Act on the clear cases; send only the grey zone to people.
ASKS
Harassment? Hate? Spam? Self-harm?
COST*
≈ 21 USD per million comments
Lead scoring
A score your sales team can read and trust.
ASKS
Fits your ideal customer? Ready to buy? Company size?
COST*
≈ 33 USD per million leads
LLM routing
Spend on the large model only when it pays.
ASKS
Which handler? How hard is the request?
COST*
≈ 27 USD per million prompts
*Estimated at the Prepaid price of 0.042 USD per million input tokens, with typical text lengths for each case. Your exact usage comes back in every response (usage.input_tokens). The Free plan covers 20 M input tokens per day.
Try one now: the playground on our site or the live demo on Hugging Face — no sign-up needed.
Your industry. Your criteria. Your model.
The public API is the general Superion S1. For decisions specific to your world, we build custom and specialized Superion models — fitted to your taxonomy, your criteria and your data, and deployed on a dedicated instance with its own capacity. Same simple API: a text in, calibrated probabilities out.
Fintech & risk
Alert triage, case review, dispute reasons, KYC notes.
Health intake
Route patient messages by topic and urgency. Decision support for your clinical team; not a medical device.
E-commerce care
Returns, refunds, delivery issues and churn signals, in your categories.
Legal & compliance
Clause flags, policy checks and review priority.
Moderation
Your community rules, your enforcement levels.
Agent routing
Your tools, your models, your definition of risky.
How we work
1 · Scope. One decision, your labels, your current process.
2 · Fit. We build the model on your criteria and examples, and measure it on your own held-out cases.
3 · Run. A dedicated instance, an immutable model ID and a contract key.
Dedicated instances and custom models start at around 2,000 USD per month. Your data is used only for your model, under a written agreement — see our privacy policy.
Talk to us about a custom model →Why a decision model, not a chat model
Calibrated
0.92 means 92%. Set the risk you accept, and audit it.
One read, many decisions
The text is billed once per request; each extra question adds only its own tokens. Output is always free.
Measured in public
#1 of 214 on S1MB English v1 (12 Nov 2026), on the benchmark's own evaluator.
Pricing
Free: 20 M input tokens per day, permanent, no card · Prepaid: 0.042 USD per million input tokens, output free · Dedicated & custom: from about 2,000 USD per month. Plan details.
Questions
Do you store my data?
No. API requests and answers are processed in memory and are not stored or used for training. See the privacy policy.
Can I run it on my own servers?
Superion S1 is offered as an API. For isolation and guaranteed capacity, we run a dedicated instance for you.
Which languages?
The public ranking is for English (S1MB English v1). For other languages, talk to us about a custom model.
Will the model change under me?
No. Each version has an immutable ID, and a version is retired only with at least 90 days notice.
How fast is it?
Well under a second per request. Use a 120-second client timeout to cover peaks.
Hand us your next thousand decisions.
Free key, no card. Or tell us about the decision you want to automate.