
South Korean financial AI company NeuroFusion has risen to the top of a benchmark testing the future-forecasting abilities of global artificial intelligence (AI) models, beating big-tech models from Google DeepMind, OpenAI, Anthropic and others.
NeuroFusion said on the 20th that its financial AI research team ranked No. 1 on the overall leaderboard of ForecastBench, a benchmark for predicting the probability of future events. As of the 17th of this month, its overall Brier Index stood at 69.4 points, the highest among the 550 models evaluated. This follows its earlier first-place finish on the preliminary dataset leaderboard in September, and it has now secured the top spot in the official rankings as well.
ForecastBench is operated by the Forecasting Research Institute (FRI), which includes Philip Tetlock, a professor at the University of Pennsylvania’s Wharton School, among its participants, and is an evaluation framework accepted to the international conference ICLR 2025. Because it compares probability forecasts for questions whose outcomes are not yet determined at the time of submission against actual results, it fundamentally prevents answer leakage or performance inflation from pre-training. The evaluation is divided into a dataset category covering economic and financial indicators such as corporate stock prices, US Treasury yields and policy rates, and a market category covering events requiring qualitative judgment, such as major-country election outcomes and whether international conflict negotiations are concluded. The overall ranking weights the two categories equally, and results are only recognized as official records after 50 days have passed since the forecast was submitted and the outcome has been confirmed.
NeuroFusion scored 65.5 points in the dataset category and 73.9 points in the market category. The gap with competing models is also clear. Google DeepMind, which submitted its model directly, trailed by 1.0 point at 68.4, while NeuroFusion widened its lead by 2.5 to 3.4 points over xAI’s “Grok 4.20” (66.9), which was evaluated by organizers with web search applied, OpenAI’s “GPT-6 Astra” (66.7), Anthropic’s “Claude Opus 5.5” (66.6) and Meta’s “Muse Spark 1.3” (66.0). Converted to a difficulty-adjusted Brier score, NeuroFusion’s “Seraphine AI” reduced forecast error by 15-16% compared with the comparison models from Anthropic and OpenAI.
Note: ForecastBench overall leaderboard as of the 17th of this month. A higher score indicates better forecasting accuracy.
It also surpassed the latest evaluation score of the “Superforecasters” (68.1 points, 9th place), a group of human expert forecasters. NeuroFusion exceeded this benchmark in both the dataset category (Superforecasters 63.4 points) and the market category (73.5 points). Among participating teams that disclosed their affiliation, only NeuroFusion and Google DeepMind scored higher than the Superforecasters.
The company cited its refined financial data infrastructure as the backdrop for this result, including “Damian,” its own four-stage dynamic time-series graph, more than 530 data pipelines built over four years, and 167 finance-specialized MCPs. The forecasting system consists of user-based hypothesis aggregation, recursive self-improvement, agentic data engineering, LLM-based backtesting and a world model.
NeuroFusion will conclude its benchmark participation with this evaluation and plans to focus on applying its research results to actual services and expanding their scope of use. It is also pushing to expand its AI investment analysis platform “Valley AI” into finance-specialized active agents.
Choi Han-chul, CEO of NeuroFusion, said, “Many AI agent services are focused on areas where they perform given tasks well, but the focus will gradually shift toward the future-forecasting domain,” adding, “If AI begins to make better judgments than humans even on problems with little data, that signals that innovation in far larger decision-making domains beyond financial markets is imminent.” He went on to say, “We will scale our current research results on a large scale and strive further toward NeuroFusion’s vision of achieving superintelligence by quantifying human thought.”
The result is seen as a signal that discussions of AI use in the financial sector are shifting from simple task automation to the realm of forecasting and judgment. Until now, competition among AI agents has centered on how accurately they perform given tasks, but with a case of surpassing human experts in an area where the answer is only confirmed after the fact—such as probability forecasting—there is a possibility that the evaluation standards for financial and investment analysis services themselves could change. However, some point out that benchmark scores are based on a set of questions submitted during a specific period, so generalization performance in real market conditions requires further verification.
Once added, BigGo Finance appears first in Google Search Top Stories, so you get the broadest, most up-to-the-minute, and most comprehensive global financial news first.
The news and data on this website are for reference only and do not constitute investment advice or an offer to buy or sell. Information is sourced from exchanges and public sources, and may be delayed, interrupted, or updated. While we strive for accuracy, we do not guarantee timeliness, correctness, or completeness. Content may include external links for which we are not responsible. By using this site, you agree that we and our partners are not liable for any losses. Investment carries full responsibility; please carefully assess risks and consult professionals. If there are errors in the content, please contact us for correction.