What is Foresight by Lightning Rod?

Foresight by Lightning Rod is an OpenAI-compatible forecasting API built for developers who need calibrated probability estimates on future events. Instead of generating plausible text like general-purpose LLMs, Foresight outputs scored, calibrated forecasts trained on real-world outcomes. It uses a proprietary Future-as-Label training method — recognized at Spotlight and the ICML 2026 AI Forecasting Workshop — and delivers benchmark-verified accuracy at a fraction of the cost of frontier models.

Who it's for

• Prediction-market developers — build bots that price contracts against calibrated probabilities and act on market edges.
• Agent builders — drop a forecasting tool into any agent workflow using the familiar OpenAI-compatible API.
• Risk and quant teams — feed news or filings to get probabilities on supply-chain shocks, policy actions, or earnings surprises.

Key features

OpenAI-compatible API with forecasting extensions

Foresight uses the same interface you already run in production. Add a base_url and api_key, and you get purpose-built forecasting features like auto-research and calibrated probabilistic answers — no new infrastructure needed.

Cheaper inference than frontier models

At $6 per 1M output tokens, Foresight costs 1.7× less than GPT-5, 2× less than Gemini 3.1 Pro, and 4.2× less than Opus 4.6. The all-in cost per 1,000 forecasts is significantly lower than any general-purpose alternative.

Future-as-Label training method

Unlike models trained to imitate generic text, Foresight learns from real-world outcomes using a method presented at the ICML 2026 AI Forecasting Workshop. This approach produces calibrated probabilities rather than confident guesses.

Built-in research and answer-type control

The API can auto-gather relevant context for any question and return a calibrated probabilistic answer. You control the answer_type and research parameters directly in the request body.

What stands out

"Frontier models make confident guesses. Foresight models return calibrated probabilities."

This is the core difference. General-purpose LLMs are optimized for plausible-sounding text, not accurate forecasts. Foresight is trained and evaluated on resolved real-world events, so its outputs are scored, benchmarked, and calibrated — not just confident. For developers building prediction-market bots, risk monitors, or decision tools, this means you get reliable probabilities you can act on, not text you have to second-guess.

Worth checking out if…

You're building any system that needs to answer "what will happen?" with a real probability — whether it's a prediction-market bot, a risk forecaster, or an agent that needs a calibrated forecasting tool. Foresight is especially useful if you're already using the OpenAI API and want to swap in a cheaper, more accurate alternative without changing your codebase.