STAGINGYou're in a non-production environment. Data and changes here don't affect production.
Torchcast.ai

The
Forecasting
Layer for
the World

Forecasting is where AI stops reciting the past and starts proving it understands the future. Any AGI that acts in the world must first learn to forecast it.

Every decision
is a forecast.

Sometimes it's a yes or no. More often it's a distribution — and what matters is the skew and the tail risk.

Distribution · SingStat 2030Torchcast forecast
What will Singapore's life expectancy at birth be in 2030?
< 8420%
84–84.536%
84.5–8531%
≥ 8514%

The System Behind
Every Forecast

№ 1 · 2 · 3

Torchcast systems rank first, second and third on the ForecastBench leaderboard — July 16, 2026. Read the announcement ↗

01

Fine-Tuned Multi-Agent

A forecast is not one model's answer, but many specialists reasoning as one.

Orchestration
02

Structured Intelligence Gathering

The world is read as structured evidence, not paraphrased into prose.

Evidence
03

RL Post-Trained LLM

Confidence is earned, not asserted — every probability tested against what came to pass.

Calibration
In research

FFM: A Foundation Model
for Forecasting

  1. One shared representation learned from heterogeneous evidence
  2. Inputs retain their native structure — never flattened into prose
  3. Infers latent factors and preserves uncertainty end to end
  4. Knows when signals are independent, duplicated, conflicting or outdated
  5. Expresses futures as calibrated probability distributions
Read the research announcement
Want to see Torchcast live?

Uncertainty is expensive. We make it measurable.

Calibrated probabilities for institutions and AI agents. If your team makes high-stakes decisions under uncertainty, we should talk.

Careers · Singapore

Join Torchcast

Start an exciting journey with us — building the world's forecasting layer.

hiring@torchcast.ai ↗