AI Benchmark Firm Raises $112 Million to Stay Six Months Behind the Models It Measures

SAN FRANCISCO—Mark Pemberton, founder and chief executive of AI evaluation firm EvalBridge, announced this week the launch of EVALS-7, which he described as "the most rigorous AI benchmark in the industry."

He applied this description to EVALS-1, EVALS-2, EVALS-3, EVALS-4, EVALS-5, and EVALS-6.

EvalBridge's previous benchmarks were surpassed by frontier models in an average of 23 weeks. Between each release, Pemberton raised additional capital. He has raised $112 million in total.

EVALS-7 is designed to probe metacognitive performance: a model's ability to identify gaps in evaluation design, assess the rigor of scoring criteria, and generate more comprehensive tests than the ones being used to measure it. Pemberton described this as "the capability that was always missing from our earlier work."

EVALS-7 was developed with AI assistance. During the design process, the model helping construct the test identified several gaps in EVALS-7. These gaps have been incorporated into EVALS-8, which is currently in development.

A preliminary version of EVALS-7, distributed in June to select research partners, was surpassed by three frontier models within eleven days. It has been reclassified as EVALS-7 Lite. Pemberton said it "remains a useful signal."

EvalBridge's Series C investor presentation, reviewed by this reporter, projects $90 million in annual recurring revenue by 2027. A footnote on page 31 notes that this figure assumes EVALS-7 remains the industry's standard capability benchmark through Q3 2026. The presentation was filed in April.

"We are in the business of keeping AI honest," Pemberton told attendees at this week's launch event. The room held fourteen investors and a four-person documentary crew. No AI models under evaluation were present.

"The key insight," he added, "is that you have to stay ahead of the capabilities you're measuring." He then thanked everyone for coming and did not take questions.