Vals raises $40M Series A from Andreessen Horowitz to expand private, task-based AI benchmarking
The AMW Read
A Series A-funded evaluation startup expanding into industry-specific and government benchmarking is an incremental update to a known player in the data-infrastructure eval space, not a debate-resolving or top-tier entrant event.
Vals raises $40M Series A from Andreessen Horowitz to expand private, task-based AI benchmarking
Vals, a San Francisco-based AI evaluation startup founded in 2024, raised a $40 million Series A led by Andreessen Horowitz last month, following a seed round led by 8VC and Bloomberg Beta a year earlier. Co-founder Rayan Krishnan, a 25-year-old former Palantir intern and Stanford AI lab alum, says the company's revenue is now eight times what it was a year ago. Headcount tripled this year from 8 to 25, with plans to add 10 to 15 more people and move to a larger office. Vals also recently launched a program offering model evaluations to federal agencies.
The pitch is a response to a real gap: public academic benchmarks lag frontier model capability and can be gamed when labs train against test sets that are openly available. Vals keeps its test material private and evaluates models on task completion in specific domains — law, finance, coding — plus newer areas including recursive self-improvement, mental health, cybersecurity, biosecurity, and compliance with the Geneva Convention. Its revenue model, charging AI companies to be tested, mirrors the College Board's SAT business and signals that independent evaluation is becoming a standalone paid layer in the AI stack rather than a cost center absorbed internally by model labs.
For builders, the shift toward private, task-specific testing means public leaderboard rankings are a weaker signal of real-world model quality than they once were. For investors, Vals's growth suggests evaluation infrastructure can generate durable revenue independent of which lab wins the underlying capability race, and the federal-agency contract points to government procurement as a distinct demand channel as AI companies — Krishnan cites Anthropic's expected IPO this year — face more formal scrutiny ahead of public filings.
#AIEvaluation #Benchmarking #AndreessenHorowitz #Vals #AIInfrastructure #SeriesA