Vals AI has introduced the Recursive Self-Improvement (RSI) Index, a standardized framework designed to measure the ability of AI models to autonomously develop their successors. The index evaluates models across five specific tasks that mirror actual AI development workflows, operating under fixed time and compute constraints to prevent brute-force results.
The scoring system uses reference points where 0 is the baseline, 0.5 represents performance levels found in published research, and 1 is the theoretical optimum. Currently, Claude Fable 5.1 is the top-performing model with a score of 35.03%, indicating that while it can execute experiments effectively, it still struggles to produce novel techniques.
The RSI Index was launched in August 2026 in collaboration with CoreWeave. This launch coincided with Vals AI securing $40 million in Series A funding, which valued the company at $400 million.
Following the funding round, Vals AI experienced significant growth over six months: revenue increased eightfold compared to the entirety of 2025, the customer base doubled, and the staff size tripled.
CEO Rayan Krishnan argues that the index provides the empirical data necessary for international governance. He notes that because development occurs within a few labs with limited visibility, a concrete metric is essential to track the trajectory toward autonomous self-improvement.