Vals Raises $40M to Scale AI Benchmarking

The big picture: Vals, a startup founded in 2024, raised $40 million in a Series A round led by Andreessen Horowitz. The company builds rigorous benchmarking systems for AI models, emerging from the observation that traditional academic benchmarks struggle to keep pace with rapidly advancing frontier models.

Why it matters:

• Proprietary Tests: Unlike many systems that rely on publicly available tests, Vals keeps its evaluation materials undisclosed to prevent companies from training models specifically to pass them, ensuring more accurate and unbiased assessments.
• Real-world Relevance: The system assesses models on their ability to complete real-world tasks across industries including law, finance, and coding, moving beyond general knowledge evaluation.
• Risk Mitigation: Vals also evaluates potential negative outcomes in critical areas such as cybersecurity, biosecurity, mental health, and recursive self-improvement.

How it works:

• Subscription Model: Companies pay for evaluations, similar to standardized tests, using the results to troubleshoot and improve their models over time.
• Rapid Growth: Vals has seen its revenue grow eightfold year-over-year and its team triple to 25 employees, with further hiring and office expansion planned.
• Government Adoption: The company recently launched a program offering model evaluations to federal agencies, anticipating a central role for such benchmarks as AI companies pursue public offerings.

The catch: The proprietary nature of Vals’ benchmarks, while preventing ‘teaching to the test,’ could also limit transparency and external scrutiny, potentially creating a black box for model evaluation. As the AI benchmarking market matures, competition from open-source alternatives and other commercial solutions will intensify.

Key Facts

• Company: Vals
• Amount: $40M
• Round: Series A
• Investors: Andreessen Horowitz (lead), 8VC, Bloomberg Beta
• Founder: Rayan Krishnan
• Announced: 2026-09-22
• Sector: AI Model Benchmarking