SAN FRANCISCO, Feb. 17, 2026 (GLOBE NEWSWIRE) — Metaculus, a forecasting platform and public benefit corporation, has launched FutureEval, a continuously updated benchmark measuring how accurately AI systems predict real-world events and how their performance compares with human forecasters.
Unlike static reasoning tests, FutureEval evaluates probabilistic forecasts about future outcomes across domains including science, technology, health, geopolitics, and AI. Because answers are not yet known, the benchmark resists test-set contamination and provides a real-world measure of reasoning.
FutureEval gives AI professionals, policymakers, and journalists clarity on how AI forecasting capabilities are evolving. It also measure AI performance against the Metaculus community and Pro Forecasters. Humans currently outperform AI models, but Metaculus says that could change soon.
“AI models are not better than the pros yet, but they’re progressing fast enough that we need to prepare for a world where they are,” says Metaculus CEO, Deger Turan.
The benchmark measures performance through three components. Its Model Leaderboard runs major AI models on Metaculus forecasting questions using a standardized prompt, scoring predictions and tracking trends over time. Current projections suggest AI systems could surpass the broader Metaculus community by April 2026 and Pro Forecasters by mid-2027.
“If AI can forecast as well as humans, that is a big deal, impacting decision-making across business, finance, marketing, law, and more,” says Metaculus AI Research Engineer Ben Wilson. “However, we need to know whether we can trust AI forecasts, and understand at what point it becomes something we can actually rely on.”
FutureEval also includes Bot Tournaments, where developers enter AI forecasting systems to compete for $175,000 in annual prizes. Human Baselines come from predictions by the Metaculus community and selected Pro Forecasters, enabling detailed comparisons; so far, Pro Forecasters have maintained a lead.
FutureEval offers several advantages over existing benchmarks. It measures decision-relevant forecasting ability, cannot be trained on directly, and avoids saturation because forecasting questions can continually increase in difficulty. The benchmark spans diverse topics and uses probabilistic scoring to evaluate accuracy, calibration, and discrimination.
FutureEval also highlights where AI and human forecasts diverge, showcasing instances when one outperforms the other. By tracking these trends, Metaculus aims to help organizations understand when AI forecasts can be trusted and how their capabilities are likely to evolve.
Contact:
Christian Williams
christian@metaculus.com
