Rayan Krishnan

Rayan Krishnan was on a path to a PhD when the post ChatGPT AI boom hit, and he made a clean break from academia to chase the one gap almost no one was focused on: independent evaluation of AI models. Instead of trying to build yet another model or a flashy consumer app, he and co founder Langston Nashold asked a more boring question that turned out to be a distribution cheat code: who actually gets to decide which models are good enough for real work.

They built Vals AI around that answer, positioning it as an independent scorekeeper that sits between frontier model builders and the enterprises that want to use those models on sensitive, high value tasks. By aiming directly at the pain of broken benchmarks and leaked test sets, Krishnan put Vals into every serious conversation about AI risk, safety, and deployment, which is exactly where the budget holders already live.

Vals AI started by attacking a very specific failure in the AI ecosystem: academic benchmarks that frontier models blow past in months, then quietly memorize as they leak into training data. Krishnan framed Vals as a neutral referee that designs private, domain specific tests, runs them on real world workflows, and gives enterprises a grounded way to compare different models, which is a far more concrete value proposition than abstract safety rhetoric.

That wedge pulled in serious capital. The company raised early funding from Pear VC and other seed backers, then closed a $40 million Series A at a $400 million valuation led by Andreessen Horowitz with existing investors like 8VC, Pear VC, Bloomberg Beta and new funds such as HRT Ventures and Next Ladder Ventures returning to the table. In public writeups Vals reports revenue growing eightfold in 2025, with the customer base doubling and the team tripling in six months, which signals that the market for independent AI benchmarking is not theory but active enterprise demand.

1. Sell picks and shovels to the model gold rush. Krishnan avoided the crowded race to build frontier models and instead built the tool everyone in that race has to use, an evaluation layer that enterprises, regulators, and even model vendors need in order to make credible performance claims. That choice put Vals in the middle of many negotiations instead of competing on the edges for end users.

2. Turn neutrality into distribution. By choosing to be an independent scorekeeper, Vals can credibly work with multiple model providers and multiple enterprise buyers at once, which converts every new integration into a demo for the rest of the market. Each time a large customer runs private benchmarks through Vals, the platform becomes the default lens that executives use to judge competing AI vendors, which quietly locks in Vals as infrastructure.

3. Package the narrative in language investors and enterprises already use. Krishnan does not talk in fuzzy AI hype, he talks about broken benchmarks, leaked test data, and real world tasks that boards and regulators care about. That framing lets Vals show up in press coverage, policy discussions, and technical buyer conversations as the sober, risk aware layer of the stack, which is exactly the narrative that converts to multi year contracts and makes a $40 million Series A at a $400 million valuation feel like a rational bet instead of a meme round.