Vals, an artificial intelligence benchmarking startup founded in 2024, has raised $40 million in Series A funding led by Andreessen Horowitz to fix what it views as a broken testing system. Existing benchmarks have become outdated and companies have learned to "game" public tests, rendering them unreliable measures of genuine model capability. With AI increasingly embedded in society, Vals argues that robust, trustworthy benchmarking is essential to verify whether models actually perform as advertised.

Unlike traditional benchmarking systems with publicly available tests, Vals keeps its test materials confidential to prevent manipulation. Rather than measuring abstract knowledge, the company evaluates models on practical, domain-specific tasks in law, finance, coding, and other industries—asking whether AI can produce work of comparable quality to humans. Co-founder Rayan Krishnan, 25, says Vals is also expanding into emerging domains including cybersecurity, biosecurity, mental health, and the law of armed conflict, examining both positive capabilities and potential negative implications if models were deployed unchecked.