Every major AI benchmark tests what a model knows or how well it answers a question. Skillprint's Signal Benchmark tests something different: whether a model can measurably improve what a real person does in the moment.
Signal, now in public beta, places AI models inside live gameplay sessions and measures whether their real-time interventions move a player toward a target mental state — greater focus, relaxation, or improvement in a specific cognitive skill. The model adjusts game parameters as the session runs; Signal records whether those adjustments had the intended effect on the person. Rankings go on a public leaderboard as session data accumulates.
The company behind Signal, Skillprint, spent six years building the underlying platform: more than 500,000 players, a library of over 1,000 games and simulations, and 150-plus mapped relationships between game mechanics and human cognitive behaviour. That foundation is what makes Signal resistant to the contamination problems that afflict static benchmarks. Each evaluation runs on a new gameplay session with a real participant; there is no fixed test set to memorise.
"Most AI benchmarks tell you how smart a model is on its own. Based on our years of decoding human cognition at scale, we've built the one that tells you whether it can actually help a human being and enhance their performance and intelligence. As AI becomes part of everyday life, we think that is one of the measures that will matter most," said Chethan Ramachandran, Co-founder and CEO of Skillprint.
At launch, Signal runs on two instrumented games, Hextris and Box Tower, across target states including Relax and Focus. No model currently leads across every measured outcome.
"We have spent years studying what gameplay can tell us about cognition and player state," said George Kachergis, Cognitive Scientist at Skillprint. "Signal reverses that question. Instead of only asking what a person's gameplay tells us about the person, we can ask what happens to that person when an AI system begins adapting the experience around them."
The benchmark is pseudonymous and consented; Skillprint does not collect personally identifiable information. The company plans to expand Signal to cover additional cognitive skills beyond the current focus and relaxation states. AI labs and researchers can join a waitlist for early access to both the leaderboard results and the underlying human-outcome data.
To stay across the latest in cloud, AI and enterprise tech analysis from Compare the Cloud, subscribe to our weekly newsletter at https://www.comparethecloud.net/newsletter
