General-purpose models often struggle with the emotional resonance required for effective B2B communication. According to an Optimizely study of 2,000 marketing leaders, more than half believe current AI tools capture basic facts but fail to connect with audiences on a deeper level. By stripping away extraneous parameters and focusing on domain-specific requirements, these new models aim to prevent the budget waste associated with using oversized generalist AI for narrow tasks.

To standardize industry performance, the company also introduced Mark-Bench, an open-source evaluation tool. It tests models against 285 tasks across 15 marketing functions. In initial benchmarks, the Optimizely Agent Platform achieved a 67% pass rate, outperforming Claude Code’s 60% while operating at half the cost. The system integrates directly with Mark-IQ, a data layer that feeds organization-specific insights—such as historical experimentation and web analytics—into every prompt.

"Marketing is a unique domain that requires specialized agent harnesses," said Shafqat Islam, President of Optimizely. "By delivering frontier quality at lower costs, our customers finally have a way to scale agentic marketing to meet real-world demands." These models are scheduled for release to all Agent Platform customers, following the company's recent launch of role-specific Virtual Teammates.