What is Frontier Labs?

Frontier model scores on the Apex benchmark jumped from 1% (o1, 12 months ago) to 40% today, a 40x improvement in one year.

“On Apex, the Frontier model right now is at about 40%. And 12 months ago, the frontier model was 01, which was scoring 1%.”

The White House voluntary frontier model evaluation process covers only cybersecurity risk, leaving all other AI risk categories unaddressed.

“Currently many of the frontier labs are working with the White House on the so-called voluntary, but a lot of industry will call that not voluntary, model evaluation. And that process and mechanism is solving only a very specific problem, and that is to manage the risk of cybersecurity. It doesn't manage anything else.”

Mercor added $300 million in net new ARR in 60 days after expanding relationships with all Frontier Labs.

“We've expanded our relationships with all of the Frontier Labs and added 300 million in net new ARR in the last 60 days.”

Frontier AI labs must continuously repurchase data for each new model because they do not retain large data stores themselves, creating a recurring revenue stream for data suppliers like Grass.

“Their customers are all the frontier labs. So, Frontier Labs need to continuously buy new data for each model because they don't store as much of it themselves.”

Harry Stebbings on calling open-source models Chinese models as a competitive scare tactic.

“Calling open-source models Chinese models is a scop by the Frontier Labs to basically trick people into thinking that they're scary and otherize them.”

Frontier model labs can rationally slash annualized inference revenue from roughly 480 billion to 120 billion by reallocating compute to training after a research breakthrough, creating extreme revenue volatility risk for public market investors.

“Let's say they have a big research breakthrough and they decide, "Wow, it is to our long-term advantage to go from eight gigs allocated to inference, two gigs allocated to training to 8 gigs on training and then your revenue just went from 480 to 120.”

Large American enterprises are afraid of both frontier lab dependence and Chinese open-source models, creating a vacuum that American open-source models could fill.

“To the enterprises that I've talked with, it is absolutely true. It's not only true that they're terrified of working with the Frontier Labs, but they're also terrified of working with the Chinese open source.”

Jack Farley on frontier AI lab IPOs forcing a reckoning between costs and revenue growth.

“Once the Frontier Labs stop juicing their numbers when they're all public, like there's going to be some come to Jesus moments for either cost is going to have to rise to make up for it or you know growth is going to have to come from somewhere.”

Well-known companies are already migrating from Western frontier AI labs to Chinese open-weight models on cost and fitness-for-purpose grounds, not just for experimentation.

“There's examples of well-known companies that have shifted from the Frontier Labs to Chinese models because they're cheaper to run. and good enough.”

Frontier Labs' FTE program is framed publicly as a sales engineer initiative but is described as actually being designed to extract client data and workflows and embed them directly into the models.

“The FTE program, people thought it was a sales engineers program. It's actually a program to extract data and workflows and implement them inside the models themselves.”