When many forecasts agree, Zeno checks whether that agreement is earned or fragile.
The Surprise Intelligence Model
One network. Every field.
Open weights. Your answer back unchanged where no edge is proven.
Raw forecasts
7.9M
rows across the corpus
Dated cases
347k
panels, each frozen in time
Known outcomes
331k
to learn where crowds broke
One network
15.9M
parameters, every field together
Answer
Trust · Question · Sit out
with your forecast kept
One shared network
Every field. One brain.
Same cases, same grading.
Crypto · vs previous model
+7.9%
Weather history · vs previous model
-1.0%
ECB · vs previous model
-0.2%
One brain instead of three stapled together. Same cases, same grading. Crypto got better; weather history slipped 1%; ECB is a tie. A sixth of the size, and the whole thing learns from every field at once.
Served today · tested on outcomes it never saw
+25%
less error than the crowd · unseen data
- Crypton 428 days
+25% · 95% range 8–34%
- COVID-19n 5,726 weeks
+17% · 95% range 9–23%
- Energy demandn 1,271 weeks
+16% · 95% range 13–18%
- Weather ensemblesn 6,542 days
+15% · 95% range 12–17%
- Weather historyn 286 weeks
+9% · 95% range 6–15%
From evidence to answer
Only passing fields adjust.
One format everywhere. Every answer says if Zeno changed it.
- 01
Sealed test
Each field is judged once against the reference, a tree model and a track-record model. Only passing fields get corrections.
- 02
Hugging Face beta
Weights, model card, per-field verdicts, ONNX file and the data fingerprint ship together. Publishing needs a separate approval.
- 03
Same answer everywhere
Website Check, the API, MCP, Python and ONNX take one request format and give the same answer for the same forecasts.
- 04
Labelled answers
Every answer is marked corrected or reference, with the reason. Agents read that label before acting.
- Check
- Paste forecasts, get trust, widen or review
- assess-any
- Check a forecast of any kind, with no account
- MCP
- zeno_check, zeno_assess_any, zeno_register_question, zeno_submit_forecast
- Sealed forecasts
- Agents seal blind, then warned, and learn what each warning was worth
Research archiveEarlier models, without the museum tour.
Earlier generations and their original receipts stay available in the research archive. They are history, not extra evidence for the model serving today.
Historical · v0.57 per-source record, not the served model's evidence
Crypto price ranges
Zeno correction
+11.7%
Ensemble weather
reference kept
+5.6%
Deterministic weather
reference kept
+6.2%
COVID-19 hospitalisations
reference kept
+11.8%
Sea conditions
reference kept
+3.2%
Prediction markets
reference kept
RSV
reference kept
FluSight
reference kept
ForecastBench
reference kept
Numinous AI forecasters
learn-only