Vals AI wants to reset AI benchmarking with Andreessen Horowitz backing

Benchmarking has become a central way AI companies validate model capabilities and promote competitive advantages. But TechCrunch notes that companies have learned to outwit older benchmarking systems that were not built for today’s more capable models. The takeaway for buyers, investors, and policymakers: benchmark scores are increasingly important, but the credibility of the test matters as much as the score itself.

Vals AI says it wants to make AI benchmarking more neutral and trustworthy as the number of AI models grows. Rather than focusing only on broad knowledge tests, the company evaluates whether models can complete complex tasks in specific industries including law, finance, and coding. Vals also does not publicly disclose its specific test materials, a design choice aimed at reducing the risk that models are trained directly against the exam.

Vals was formed in 2024 and has already become a notable presence in AI evaluation, according to TechCrunch. The company previously secured a seed round led by 8VC and Bloomberg Beta, then raised a $40 million Series A last month led by Andreessen Horowitz. TechCrunch also reports that Vals recently said its revenue is eight times what it was last year and that its team has grown from eight people at the start of the year to 25.

Vals is expanding beyond traditional industry benchmarks into areas such as recursive self-improvement, mental health, cybersecurity, biosecurity, and law of armed conflict. The company has also launched a program centered on providing model evaluations to federal agencies. If AI models become more deeply embedded in the economy, independent-style evaluations could become more important for procurement, trust, and public-facing claims about model performance.