Optimizely introduced a family of purpose-built, post-trained AI models for marketing on September 1, 2026, designed to match frontier-level quality at roughly one-tenth the cost of general-purpose large language models. The move targets a persistent tension in marketing teams: generalist AI models carry excess parameters and token costs that add up fast when applied to domain-specific work like campaign copy, experiment analysis, and customer behavior interpretation.
The company's research underscores the gap. A global study of more than 2,000 B2B marketing leaders found that 53% say current AI tools can capture the facts of a brand but struggle to capture the emotional resonance that helps it connect with audiences. Optimizely's models were built around a different premise - AI can take on more of the work without stripping away the context and understanding that make marketing distinctive.
Post-training as a cost-quality strategy
Frontier models are built to be generalists, capable of a wide range of tasks, but that breadth comes at a cost. Marketing tasks are nuanced and domain-specific, so a general model carries more parameters, pulls in extra context, and spends tokens it never needs. Optimizely's approach strips away the excess from state-of-the-art models, leaving models built specifically for marketing tasks.
"The initial results of our AI Lab have surpassed our expectations," said Imran Yousuf, Head of AI Engineering at Optimizely. "We believe post-training models is the long-term solution to the challenging cost vs. quality debate. Our purpose-built family of models is built to get work done fast and cost-effectively, without sacrificing quality."
Mark-Bench and brand-aware context
Alongside the models, Optimizely introduced Mark-Bench, an open source benchmark built to evaluate AI performance for the marketing domain. It tests models against 285 tasks spanning 15 marketing functions and over 6,000 criteria, including writing a press release, creating a social post, and writing email copy. The company's intent is for marketers, researchers, and other AI providers to test their own models and agentic harnesses against a shared, objective yardstick that measures both cost and performance on marketing-specific work.
The models draw on Mark-IQ, the Agent Platform's data layer, which is built on each organization's own context - including experimentation history and web analytics. That means every model has the brand's context without a marketer having to rebuild it with every prompt. As evaluated by Mark-Bench's all-pass rate on default configurations, the Optimizely Agent Platform scored 67% compared to 60% for Claude Code at 2x lower cost.
"The problem with generic harnesses is that they're good for general use and productivity but very inefficient when it comes to domain-specific work," said Shafqat Islam, President of Optimizely. "Marketing is a unique domain that requires agent harnesses to be specialized. By delivering frontier quality at lower costs, our customers finally have a way to scale agentic marketing to meet real-world demands rather than getting stuck in pilot purgatory."
Why this matters for marketing professionals
For marketing teams, model selection directly shapes budget and output quality. Defaulting to one general-purpose model for every task inflates costs as agentic work scales. Purpose-built models that understand brand context - without requiring marketers to re-explain that context in every prompt - shift the calculus from "what can AI do" to "what should AI do next." The practical takeaway is that AI for Marketing is moving toward specialized tooling that fits into existing workflows rather than demanding new ones. For managers building team capabilities, an AI Learning Path for Marketing Managers can help bridge the gap between understanding these tools and putting them to work on real campaigns.
Your membership also unlocks:
