AI model evaluation company Vals AI completes $40 million Series A funding, valued at $400 million, led by a16z
Vals AI completes $40 million Series A funding
AI model evaluation company Vals AI has completed $40 million in Series A funding, with a valuation of $400 million, led by a16z.
Existing investors such as 8VC, Pear VC, and Bloomberg Beta continue to participate.
Vals specializes in third-party evaluation of large models, testing different models using real tasks from programming, finance, law, and healthcare. Its results have been referenced in model cards from OpenAI, Anthropic, Google, Meta, and xAI, with revenue this year reaching eight times that of the entire year of 2025.
a16z values the role of third-party evaluation, as public benchmarks are increasingly susceptible to score inflation, training data contamination, or targeted optimizations by vendors.
Vals has fully opened Vals Smith, which allows any GitHub repository to be turned into a Coding Benchmark, extracting real development tasks from historical PRs and using hidden tests to check model performance.
Market-wise, the demand for independent evaluation drives capital towards trust-layer infrastructure, benefiting enterprise buyers with real task data that reduces selection risks, while model vendors face pressure to improve actual work performance rather than just leaderboard scores.
Source: Public information
ABAB AI Insight
Vals was founded by Rayan Krishnan and Langston Nashold, who have backgrounds from Stanford. The team size has tripled in six months, and the customer base has doubled, indicating a rapid shift from experimental tools to critical enterprise infrastructure.
In terms of capital, a16z led the investment and brought in new shareholders with quantitative backgrounds like HRT. The funds will be used to expand real workflow benchmarks and risk assessments, motivated by the goal of shifting model selection from self-reported scores to independently verifiable standards, thereby capturing the trust premium in AI procurement decisions.
Similar cases can be seen in Moody’s independent scoring role in the credit market, as well as early Benchmark companies that quickly commercialized after being referenced by major firms; Vals is currently in an expansion phase transitioning from public leaderboards to customized enterprise evaluations.
The structural judgment indicates a transfer of pricing power: third-party real task evaluations are beginning to replace vendor self-assessments, as buyers need external judges under conditions of information asymmetry, thereby reconstructing the trust and procurement logic in the model market.
ABAB News · Cognitive Law
Self-reported scores are never as reliable as third-party judges
Real tasks are the ultimate test for models
The trust layer determines the final premium in AI procurement