Vals Secures $40 Million Series A as AI Benchmarking Revenue Jumps 8-Fold
Updated
Updated · TechCrunch · Sep 19
Vals Secures $40 Million Series A as AI Benchmarking Revenue Jumps 8-Fold
3 articles · Updated · TechCrunch · Sep 19
Summary
$40 million in Series A funding led by Andreessen Horowitz gives Vals fresh capital after less than two years in business and a seed round last year.
Vals is pitching itself as a fix for outdated AI benchmarks by using undisclosed tests and industry-specific evaluations in law, finance and coding, aiming to measure real-world performance and risks rather than memorized answers.
Revenue is now 8 times last year's level, and the startup has expanded from 8 employees at the start of 2026 to 25, with plans to add 10 to 15 more and move to a larger office.
The company has also launched a federal-agency evaluation program, betting that tougher model testing will become more important as AI systems spread through the economy and more AI companies head toward public markets.
Could commercializing AI evaluations create a monopoly on trust where only wealthy tech giants can afford to prove their models are safe?
With frontier models failing half of real-world tasks, are companies rushing to deploy AI systems that remain fundamentally unready for enterprise use?
If AI benchmarks are kept entirely secret to prevent cheating, who audits the evaluators for hidden biases or flawed metrics?