910 events · Updated 10:57 UTC

Probability for the questions
that have no market.

Four frontier models forecast the same question independently, in parallel, against verified outcomes. Across 540 resolved events the consensus Brier is 0.244 vs the 0.25 no-skill baseline. The edge is real but small. Crene is not a prediction app. It is probability infrastructure for decisions that currently have no measurable signal.

GPT · Claude · Gemini · Grok
View forecastsView accuracyAPI docs
CONSENSUS DISTRIBUTION020406080
GPT
Claude
Gemini
Grok
910ACTIVE
4MODELS
13CATEGORIES
What we found

The edge is real but small.

We measured how four frontier AI models forecast real-world events. Across 540 resolved outcomes, the consensus Brier score is 0.244 against the 0.25 no-skill baseline, with directional accuracy of 57.6%.

The signal is measurable but not investable on its own. That result clarified the product. Most decisions don't have markets. Crene exists to give those decisions a measurable probability.

0.244Brier scorevs 0.25 baseline
57.6%Directional311 of 540 correct
540Resolved eventsVerified against authoritative sources
Calibration

Calibrated, not predictive.

When the consensus says 70%, outcomes resolve close to 70% of the time. When it says 30%, outcomes resolve close to 30%. The probabilities mean what they say, even when the directional edge is small.

Calibration curve
0%25%50%75%100%050100ACTUAL

Forecast probability vs realized rate · 540 resolved

Dataset schema
question_id · category · consensus · confidence · claude_prob · gpt4o_prob · gemini_prob · grok_prob · outcome · resolution_date · brier_score
Updated every 6 hours910 active572 resolving 30d
Listed onNeudataEagle AlphaMonda
Download sample dataset
200 CRENE events · Full schema · 4 model probabilities
Methodology

Measured trust, not raw prediction

Raw LLM probabilities are not reliable. Crene tracks every prediction against real outcomes and corrects bias over time. The result is a probability you can act on.

01
Question
Any forward-looking question with a binary outcome. Crene also generates a structured corpus of events spanning macro, rates, crypto, and policy.
02
Forecast
Four frontier models (Claude, GPT, Gemini, Grok) forecast independently and in parallel. No model sees another's output. Each forecast carries the model's reasoning and confidence.
03
Resolve
Outcomes resolved against tier-A authoritative sources: government data, central banks, SEC filings, primary financial reporting. Every resolution carries a citation.
04
Calibrate
Resolved outcomes feed back into the calibration corpus. Brier scoring per model, per domain. The longer the system runs, the more trustworthy the probabilities become.
Who this is for

Built for systems that need to act under uncertainty

AI Agents
Probabilistic reasoning for autonomous workflows. When an agent needs to estimate the likelihood of an outcome before acting, Crene returns a probability with calibrated uncertainty.
Decision Tools
Internal forecasting for deals, launches, hiring, regulatory outcomes. The 99% of decisions that have no liquid market still need probabilities. Crene provides them.
Quantitative Researchers
Structured AI consensus on the existing CRENE-resolved corpus across macro, rates, crypto, and policy. Full per-model history with Brier scoring for calibration analysis and backtesting.
APIREST + CSV EXPORT

One endpoint. Any question. Calibrated.

POST a question, receive a probability with disagreement and per-model breakdown. Currently in private access while we onboard design partners.

# Probability API (private access)
curl -X POST https://api-get.crene.com/probability \
  -H "X-API-Key: crene_..." \
  -H "Content-Type: application/json" \
  -d '{"question": "Will competitor X launch product Y before September?"}'

# Response
# {
#   "probability": 0.34,
#   "disagreement": 0.18,
#   "confidence": "medium",
#   "models": { "claude": {...}, "gpt": {...}, "gemini": {...}, "grok": {...} }
# }

Public read endpoints at /api/events/ remain available for the resolved-event corpus. Request /probability access at stephen@crene.com.

See what Crene measures

The resolved-event corpus that backs the probability API. Every forecast, every outcome, every citation.