Vals AI

The page compares AI models and large language models across coding, finance, legal, and science benchmarks, including independent leaderboards, evaluation costs, and testing methodology.

Independent AI evaluation for government. Explore Vals research on public services, cybersecurity, youth safety, national security, and environmental impacts.

Archived AIME results for AI models are available, comparing mathematics accuracy, evaluation costs, and Vals AI’s testing methodology.

Alexi scored 80% accuracy, 75% authoritativeness, and 70% appropriateness in VLAIR's 200-question legal research study, versus a lawyer baseline.

How Vals AI evaluates models and agents: task design, scoring, costs, latency and uncertainty across public and private benchmarks.

Vals AI describes how it developed the Vibe Code Bench to evaluate coding agents on web applications, detailing the benchmark design, testing approach, and lessons learned.

This page compares AI writing detectors and large language models on private human and AI-generated text, analyzing detection accuracy, false positives, and adversarial rewrites.

Research, benchmark deep dives, and evaluation findings from the Vals AI team.

Media coverage of Vals AI benchmarks, research, and model evaluations.

News, benchmark releases, research, and updates from Vals AI.

Browse AI model comparisons with shared benchmark scores, reported uncertainty, cost per test and latency. Find a pair to compare on Vals AI.

CoCounsel's VLAIR scores across four legal tasks: Document Q&A, Chronology Generation, Document Summarization, and Data Extraction, vs a lawyer baseline.

Choose AI models and benchmarks to compare performance, token pricing, cost per test and latency in Vals AI's interactive comparison tool.

Counsel Stack scored 81% accuracy, 77% authoritativeness, and 71% appropriateness in VLAIR's 200-question legal research study, versus a lawyer baseline.

Vals AI is an independent evaluator of artificial intelligence, building benchmarks and evaluation infrastructure for real-world work.

Vals AI independently tests AI models on real-world tasks, presenting benchmark results, model comparisons, and research across coding, finance, and law.

Guidelines for responsibly reporting security vulnerabilities affecting Vals AI systems and services.