Lightning Rod - Calibrated forecasting API for developers building prediction tools

What it does

Lightning Rod offers Foresight, an OpenAI-compatible forecasting API that returns calibrated probabilities for questions about future events. Unlike general-purpose LLMs that produce plausible text, Foresight models are trained using a proprietary Future-as-Label method and evaluated on real-world outcomes. The API supports features like auto-research and answer type selection, and is designed to be a drop-in replacement for standard chat completions in forecasting workflows.

Who it is for

Foresight is built for developers creating forecasting agents, prediction bots, market makers, risk forecasters, quant signal trackers, and event monitors. It targets enterprise, government, and startup teams that need reliable, calibrated probability estimates rather than confident guesses from frontier models.

Why it matters

General-purpose LLMs are not optimized for forecasting; they imitate text and can be expensive in agentic loops. Foresight claims higher accuracy (measured by Brier Skill Score on resolved Polymarket questions) at a fraction of the cost. For example, Foresight v4 (Full) costs $6 per 1M output tokens versus $10–$25 for frontier models, and achieves a Brier Skill Score of 26% compared to 13–19% for alternatives. This makes calibrated forecasting more accessible for automated decision-making.

Launch signal

Foresight v4 is the latest model release, highlighted as "New" on the website. The company has published research and benchmarks, and the API is available now with documentation and a Python client example. The website mentions trust from enterprise, government, and startup customers, though no specific names are provided.

Brand and naming

The name "Lightning Rod" suggests attracting and channeling powerful forces (like lightning) into a controlled path, metaphorically capturing the idea of focusing predictive power into calibrated outputs. "Foresight" directly conveys the product's purpose of seeing future events. The branding emphasizes precision and reliability over the confident but uncalibrated outputs of frontier models.