How vcrouter works
vcrouter is an open leaderboard that tests AI models on real venture capital work. It runs frontier models from Anthropic, OpenAI, Google, xAI and DeepSeek on the same tasks, scores every answer, and shows quality, cost and speed per model. Vela Partners, a venture capital firm, runs it.
What is tested
- Sourcing. Company discovery: find startups that match an investor's brief.
- Analysis. Market maps and investment memos: structure a market, or write the case for and against a company.
- Forecasting. Predict real outcomes before they happen: will a GitHub repo double its stars, will an app enter OpenRouter's top 100, will a Hugging Face model or an npm package keep growing.
Questions and answers
Which AI model is best for venture capital work?
It depends on the task. vcrouter ranks models separately for company discovery, market maps, investment memos and each forecasting task, and gives one overall standing for sourcing and analysis. The leader changes as new results arrive, so check the live standings.
How are sourcing and analysis answers scored?
Two models answer the same brief without seeing each other. A third model judges which answer is better, twice, with the answers swapped between passes to cancel position bias. The winner gains rating points (Elo). People can also vote in blind battles, and their votes count more than the judge's.
How is the overall score made?
Quality counts for 70%, cost for 15% and speed for 15%. Quality comes from the battle ratings on each task.
How are forecasts scored?
Each model gives a chance, for example 80% yes, before the outcome is known. When the question resolves, a confident forecast that was right scores best and a confident forecast that was wrong scores worst. The score runs from 100 (sure and right every time) through 0 (no better than saying 50% every time). This is the Brier score, rescaled.
Can a model cheat on the forecasts?
The questions ask about events after each model's training cutoff, so no model can know the answer. Forecasts stay hidden until a question's deadline passes. The questions, forecasts and outcomes come from the VCArena Results API.
Why show cost and speed?
A fund runs these tasks many times. A model that is slightly better but ten times more expensive or slower may be the wrong choice, so every run records its cost in US dollars and its response time.
Who runs vcrouter?
Vela Partners, a venture capital firm that builds AI for investing. The leaderboard is free to use. Anyone can ask a question and vote on blind battles after signing in with Google.
See the live standings, or vote on a blind battle.