Foresight by Lightning Rod

Foresight by Lightning Rod is a specialized forecasting API for developers building prediction-market bots, agentic decision tools, and risk forecasters. Unlike general-purpose LLMs that produce plausible text, Foresight returns calibrated probabilities for any future event question. The API is drop-in compatible with OpenAI's chat completions endpoint, so developers can switch from GPT-5 or other models with minimal code changes. Foresight is trained with the Future-as-Label method, which learns directly from real-world outcomes rather than human labels, and has been peer-reviewed. The models are benchmarked on live prediction markets like Polymarket and ProphetArena, consistently outperforming frontier models (GPT-5, Gemini 3.1 Pro, Opus 4.6) in Brier Skill Score while being 1.7x to 4.2x cheaper per 1M output tokens. Foresight v4 offers two variants: Low (public) and Full (contact). The API supports additional parameters like 'research' for auto-gathering context and 'answer_type' for calibrated probabilities. A recent publication demonstrates forecasting for clinical events from raw notes. Foresight is ideal for quant traders, AI agent builders, enterprise risk teams, and government agencies, but not for general chatbot or content generation tasks.