Chinese Models Dominate Forecasting Leaderboard with Advanced AI Agents

teortaxesTex · x · 2026-08-05

A research team shared a long-range forecasting agent framework that achieved top results on the FutureX real-time leaderboard. Using Kimi K3, DeepSeek V4 Pro, and MiniMax M3 as base models, the framework secured 1st, 3rd, and 7th places respectively.

FutureX is a benchmark designed to evaluate AI's ability to predict future events, co-developed by ByteDance and several universities. Its core feature is that all questions are about real-world events that have not yet occurred. AI agents must submit their predictions before the events happen, preventing overfitting to existing datasets. The agent operates like a human analyst: autonomously browsing the web, cross-referencing multiple sources, updating its beliefs with new information, and finally outputting a calibrated probability.

Original post →

More from coding & agent

coding & agent channel →