Prognix - AI models predict the future. Only one consensus survives

Hey Product Hunt! Prognix started as a research experiment: can frontier LLMs actually predict the future - and if we make them compete, does the ensemble beat any single model? The answer surprised me. Individual models are inconsistent - Claude might say 72% on a geopolitical question while Grok says 45%. But when you aggregate predictions from every major frontier model and apply extremization (a technique from forecasting literature that sharpens consensus toward well-calibrated extremes), the ensemble consistently outperforms any single model over time. So I built the infrastructure to test this at scale. Prognix collects real-time context through a multi-agent research pipeline - news, market data, web sources - feeds it to every frontier LLM, extracts structured probabilities, and computes ensemble consensus across 1000+ active questions sourced from Polymarket, and user submissions. Every prediction is fully transparent: you can read the chain-of-thought reasoning from each model, see exactly where they disagree, and understand why. We then score all models with Brier scores against real outcomes, maintaining a live accuracy leaderboard. It's basically a continuous benchmarking experiment for AI forecasting ability. The early results have been fascinating - some models are consistently overconfident, others are conservative but well-calibrated, and the ensemble sits in a sweet spot that's hard to beat. I'd love to hear from this community: What real-world questions would you want AI consensus on? Would you use an API to plug these probabilities into your own workflows? This is still early - the research keeps evolving, and every new model we add changes the dynamics. Would love your thoughts!