Vals raises $40 million for AI model evaluation — TechCrunch
In San Francisco, United States, Vals, a startup developing evaluation systems for artificial intelligence models, raised $40 million last month in a Series A round led by Andreessen Horowitz. The company, founded in 2024, aims to improve industry-specific systems for assessing the capabilities of AI models.
As TechCrunch reports, Vals had previously also received seed funding in a round led by 8VC and Bloomberg Beta. Company co-founder Ryan Krishnan believes that academic benchmarks are not keeping pace with the rapid development of new AI models and do not always make it possible to verify whether their practical capabilities match developers' claims.
Closed tests for industry tasks
Vals does not disclose the specific materials used in its tests. According to the company's plan, this should make it more difficult to prepare models directly for evaluation tasks. The startup focuses not only on systems' general knowledge, but also on whether they can perform complex tasks in specific fields, including law, finance, and programming.
More current news is available on the UA.News Telegram channel Telegram.
The company also tests potential negative consequences of using models. Areas Vals is working on include recursive AI self-improvement, mental health, cybersecurity, biosecurity, and international humanitarian law, including the application of provisions of the Geneva Conventions.
Team growth and work with agencies
Companies pay Vals to test their models in order to identify problems and improve their systems. According to Krishnan, such evaluations are becoming one of the factors in selecting AI models for deployment.
Vals said its revenue is now eight times higher than last year. Since the beginning of the year, its staff has grown from eight to 25 employees. The company plans to move to a larger office and hire another 10 to 15 people. It has also launched a model evaluation program for federal agencies.
