AI companies have mastered the art of gaming legacy benchmarks. When metrics look good, PR follows. But these older testing frameworks weren’t built for modern models, creating a credibility gap that Vals aims to close. The San Francisco startup just secured $40 million in Series A funding led by Andreessen Horowitz, with participation from 8VC and Bloomberg Beta. This capital injection follows a seed round raised last year, signaling strong investor confidence in their mission to overhaul how we measure machine intelligence.
Rayan Krishnan, the 25-year-old co-founder, spotted the flaw early. Formerly an intern at Palantir and a researcher at Stanford’s AI lab, he noticed academic benchmarks lagging behind rapid industry advances. “We were seeing a bunch of new, very capable models come to market quickly, and the academic benchmarks [were] not keeping up with that frontier advance,” Krishnan says. Traditional tests often rely on public datasets, allowing developers to train models specifically to pass those exams. Vals flips this script by keeping its test materials private. Instead of abstract knowledge checks like bar exams, they evaluate real-world task completion in law, finance, and coding.
The approach goes beyond simple accuracy. Vals analyzes negative implications, asking what happens if models run wild. Their scope now includes recursive self-improvement, mental health, cybersecurity, and even the Geneva Convention. Companies pay for these rigorous evaluations much like students pay for SATs. It sounds counterintuitive to pay for potential bad news, but Krishnan argues it helps firms troubleshoot and improve. The market agrees: Vals revenue is eight times higher than last year, and the team has tripled to 25 employees.
This shift isn’t just about technical validation; it’s becoming a financial imperative. As AI giants like Anthropic and potentially OpenAI prepare for public offerings, reliable benchmarks will drive investor trust and regulatory compliance. Krishnan predicts these evaluations will become central to public filings and investment decisions. With federal agencies now adopting their program, Vals is positioning itself as the gatekeeper of AI credibility. If accurate measurement becomes the currency of trust, this startup may well define the standard for the next decade of AI development.
