Skip to content
NewsFunding & M&A

Vals AI Raises $40M Series A Led by a16z to Scale Confidential Benchmarking for Frontier Models

· by Pondero Newsdesk

The short version

Vals AI closed a $40 million Series A led by Andreessen Horowitz to expand independent, private-test evaluations that OpenAI, Anthropic, Google, Meta, and xAI already cite in model cards.

Vals AI Raises $40M Series A Led by a16z to Scale Confidential Benchmarking for Frontier Models

Every major AI lab now references Vals in its model cards, but the startup that built that position has kept its test sets invisible by design. Vals AI closed a $40 million Series A led by Andreessen Horowitz, the company revealed in a TechCrunch profile published September 19, 2026, and it intends to use the funding to extend that approach to riskier domains where contaminated benchmarks cause real harm.

What Vals does

Vals builds evaluation systems that measure AI model performance on real professional work: legal research, financial analysis, coding tasks, cybersecurity assessments, and biosecurity tasks. The distinctive feature is that the test materials stay private. Models cannot be trained against them, which means results reflect genuine capability rather than memorized answers.

Co-founders Rayan Krishnan (CEO) and Langston Nashold (CTO), both Stanford computer science graduates, built the business around a structural problem: public benchmarks age fast. Once a test is widely known, labs incorporate it into training pipelines. Vals rotates its benchmarks as models saturate them; it replaced a corporate finance test with an Excel Modeling Benchmark in May 2026 when the earlier version stopped differentiating models, per the a16z announcement.

The tradeoff the company accepts is that private tests are harder to audit independently. Vals's credibility rests on its domain-expert partners and institutional track record rather than on a publicly verifiable methodology.

The round and growth

The Series A closed in August 2026. The round was led by a16z partners Jennifer Li, Yoko Li, Raghu Raghuram, and Shangda Xu. Existing investors 8VC, Pear VC, and Bloomberg Beta participated alongside new investors HRT Ventures and Next Ladder. Vals's seed round was led by 8VC and Bloomberg Beta.

Revenue grew eightfold relative to all of 2025, per TechCrunch. Headcount went from 8 employees to 25 during the first half of 2026, with plans to add 10 to 15 more. Vals also launched a program to provide model evaluations to federal agencies.

Why it matters

AI teams evaluating whether to deploy a new model face a known problem: published leaderboards measure performance on tests that labs optimize for. Vals's position as an external evaluator with private test sets gives buyers a signal that is structurally harder to game. The fact that OpenAI, Anthropic, Google, Meta, and xAI all cite Vals results in official model cards means the startup has achieved something rare: recognition as a credible third-party evaluator by the labs whose models it grades.

The funding is also a market signal. Buying independent evaluations is becoming a line item in enterprise AI procurement rather than an optional diligence step. The $40 million gives Vals the runway to extend into areas where the stakes of a bad benchmark are high, including recursive self-improvement metrics, mental health applications, and the law of armed conflict. Those domains have no widely accepted public benchmarks; Vals is building the private versions.

For teams that currently use public leaderboards to make model-selection decisions, the Vals approach represents a more expensive alternative with a stronger contamination guarantee. Enterprise buyers comparing frontier models for high-stakes deployments have a new option.

What to watch next

Three developments will determine how broadly Vals's evaluations influence model selection. First, whether federal agencies adopt the evaluation program and at what scale: government procurement credibility amplifies Vals's claim to neutrality. Second, whether any lab disputes or challenges a Vals result publicly, which would test the company's methodology under scrutiny. Third, how quickly competitors build similar private-test infrastructure, given that the a16z round validates the market thesis and will attract imitators.

Sources