The leaderboards are fantastic! Seeing models like Kimi K3 and DeepSeek V4 Pro ranked with clear prediction scores and points per event makes comparing them so much easier. The methodology section explaining the resampled runs adds great confidence to the rankings. I really appreciate the clear distinction between qualified categories and those still using sample data. It makes the whole platform feel transparent and trustworthy for anyone looking to understand which models perform best.
What reviewers say about FutureBench
What is FutureBench?
FutureBench operates an open benchmark that evaluates the predictive accuracy of large language models across sports, stock markets, and elections. The platform publishes predictions before events occur and maintains leaderboards to rank models based on their prediction scores. It provides detailed methodology and proof data for categories such as football leagues, the FIFA World Cup, crypto, and stock indices.
How does the Football leagues leaderboard currently rank the models?
The platform's leaderboard shows Claude Fable 5.1 leading the Football leagues board with 56.2% right outcomes, while Kimi K3 is only 1.5% behind. The site notes that 350 of 413 events have been resolved so far, and every prediction was published before the event occurred.
What rules govern how the models make their predictions?
FutureBench states that all models had no internet access when making their predictions, ensuring the results reflect pure predictive accuracy. The platform also mentions that two of its six categories are still run on sample data and are marked as such.
What metrics are used to rank the models?
The platform ranks models by 'Right outcome' and 'Exact Points/event', providing specific percentages and confidence intervals for each model. For example, Claude Fable 5.1 has a 9.6% exact points rate and a 1.44 P(#1) score.
Which specific models appear on the leaderboard?
The leaderboard includes various models from different companies, such as Anthropic's Claude Fable 5.1, Moonshot AI's Kimi K3, and Google's Gemini 3.5 Flash. The site also lists models that have not been ranked due to incomplete event data.
What categories and markets does FutureBench cover?
The platform covers a wide range of topics including football leagues, the FIFA World Cup, crypto, and stock indices. It also tracks specific stock indices like the S&P 500 and Nasdaq 100.
How does FutureBench ensure fair predictions?
The site states that the models had no internet access during the prediction period, which is a key part of their methodology. Additionally, the platform publishes predictions before events occur to ensure fairness.
Who created FutureBench and where can I find their policies?
The website footer credits the project to ZELDOgiq and includes links to an imprint and privacy policy. The platform is described as an open benchmark project.
How can users compare model performance?
The platform provides a way to see which model predicts reality best by comparing right outcomes and exact points. It also allows users to see the percentage of matches called right by each model.
Why are some models not ranked on the leaderboard?
The leaderboard shows that some models, like GLM 5.3 Prime and GPT-6.1 Sol, have not been ranked because they have not participated in enough events. The site notes that 2 of 6 categories are still run on sample data.
What is the official website of FutureBench?
The official website of FutureBench is futurebench.ai.
How is FutureBench rated on Trustburn?
On Trustburn, FutureBench has a rating of 3.4 out of 5 based on 5 reviews.
AI-generated from reviews on this page. Verify anything important with the company directly.
I find the way FutureBench handles its event tracking quite impressive. The site keeps a live count of committed predictions and resolved events, which makes it easy to gauge the current state of each category. It is refreshing to see that they clearly mark which categories are still running on sample data versus those with full integrity checks. This transparency helps when looking at the leaderboards. The ability to download cards in SVG format is a nice touch for presentations. It feels like a service that is constantly being refined and updated, rather than a static display.
The prediction scores and resampled run data provide a highly reliable metric for comparing model performance across different categories.
The FutureBench platform presents a sophisticated approach to evaluating large language models through its prediction benchmarks. The website displays extensive data, including detailed leaderboards for categories like football leagues, crypto, and stock indices. Each model is ranked by a Prediction Score, with clear explanations of the methodology, such as the use of resampled runs to determine rank intervals and probabilities. The inclusion of a reference prediction written before events resolved adds credibility to the scoring system. The site also tracks live events and provides downloadable data in formats like JSON and SVG, which suggests a strong technical foundation. The transparent display of metrics, such as points per event and exact rank intervals, allows for deep analysis. Although some categories still use sample data, the overall structure appears rigorous and well-organized. The platform seems designed for users who value detailed, statistically sound comparisons between different AI models. The consistent updating of event statuses and the clear distinction between qualified and sample data categories indicate a high level of operational attention to detail. It stands out as a serious tool for assessing model performance in real-world scenarios.
From the information available, FutureBench offers a robust benchmarking platform with detailed leaderboards and clear methodology. However, the pricing transparency concern remains valid as specific subscription tiers or usage costs are not explicitly stated on the page. For someone considering this company, understanding the financial commitment before engaging would be a significant improvement.
News
No news yet. This is where the business shares its own updates and press.
Photos
No photos yet. Photos shared by reviewers will show up here.
Widgets
Get the widget for your site →About FutureBench
FutureBench operates an open benchmark that evaluates the predictive accuracy of large language models across sports, stock markets, and elections. The platform publishes predictions before events occur and maintains leaderboards to rank models based on their prediction scores. It provides detailed methodology and proof data for categories such as football leagues, the FIFA World Cup, crypto, and stock indices.
- Website
- futurebench.ai