Vals AI

Overview

Vals AI is an independent AI evaluation company building real-world professional benchmarks (law, finance, engineering, medicine) with private test sets and automated expert-grade grading. Positions as the “trust layer” between frontier labs and enterprise buyers.

Recent Developments

• 2026-09-17: Legal Research Bench cited by openai for astra-for-law — 54.0% vs 38.7% correctness vs GPT-6 Astra + web search (2026-09-17-openai-astra-for-law)
• 2026-08-13: $40MSeriesAat$400M valuation — andreessen-horowitz lead; 8VC, Bloomberg Beta, HRT Ventures (2026-08-14-vals-ai-40m-series-a)
• Revenue 8x in 2025; customer base doubled; team tripled in 6 months
• Launched: Vals Smith (GitHub repo benchmarks), frontier-risk benchmarks (cyber, mental health), Vals Index 2.0
• Results cited in model cards from openai, anthropic, google, meta, xAI

Related

• ai-benchmarks
• frontier-model-economics
• building-to-the-test
• andreessen-horowitz

Sources

• 2026-08-14-vals-ai-40m-series-a
• 2026-08-13-vals-ai-a16z-primary
• 2026-08-13-vals-ai-series-a-blog