Vals AI is an independent AI evaluation company that measures how well leading models perform on economically valuable, real-world tasks. Its benchmarks span finance, software, cybersecurity, healthcare, legal work, and frontier risks, including recursive self-improvement and mental health. Vals AI runs its own evaluations and develops many benchmarks in-house, aiming to provide unbiased comparisons that help customers understand which models and tools best suit their needs.

The company’s benchmark suite assesses performance alongside practical considerations such as cost, latency, and token usage. Its offerings include the Vals Index, specialized evaluations such as Finance Agent Bench, CyberBench, LegalBench, and software benchmarks, as well as model comparisons and detailed evaluation reports. Vals AI also publishes research and analysis on emerging topics, including AI writing detectors and web search quality. For its Web Search Index, it works with search providers while independently testing them under consistent conditions, using expert-written legal and finance tasks.