VCBench: VC-focused Model Benchmarks

An open platform to benchmark precision of AI models for venture capital.

Seed Precision Benchmarks

The $100K and $3M indexes are investing strategies of investing in any startup that raised more than $100K and $3M, respectively.

Y Combinator and Tier-1 VCs are human pickers. The precision at pre-seed, angel and seed stage financing vary between 3%-6% for best human pickers.

For the Vela AI model (GPTree), the only input is the founder's background.

Methodology

This model aims to predict the success of startups based on the founding team's background, focusing solely on instances where the model makes a decision by analyzing a single founder's background. Our analysis considers inception-stage investments, which are earlier than seed-stage, with precision calculated using real-world outcomes such as $500M+ M&A, $500M+ IPO valuations and $500M+ raises.

Vela AI's model (GPTree) focuses specifically on evaluating founder characteristics, employing sophisticated machine learning and LLM techniques to measure the likelihood of achieving outlier success. These calculations are based on historical data, benchmarked against well-known accelerators and venture capitalists for comparison.

Submit Your Model

We encourage researchers and other VC firms to submit their models for evaluation. Please contact us at benchmark@vela.partners with details about your model, including methodology and performance metrics.