The Human Creativity Benchmark of Contra Labs distinguishes convergence and divergence in the evaluation of Generative AI.

Contra Labs’ Human Creativity Benchmark evaluates generative AI, showing Claude Opus 4.6 leads in ideation while Gemini 3.1 Pro dominates mockup stages.

First report published by Contra Labs Research on April 30, 2026, the Human Creativity Benchmark (HCB) offers a framework for evaluating generative AI models based on the criteria actually applied by professional creatives. The study leverages the Contra network, which brings together over 1.5 million freelancers who have billed over $250 million through the platform.

The protocol separates two signals that traditional benchmarks conflate. Convergence refers to the dimensions where evaluators agree (typographic legibility, call to action placement, visual hierarchy, layout consistency) and pertains to teachable best practices. Divergence refers to those where they legitimately disagree (art direction, mood, conceptual risk) and relates to taste, and thus to the model's steerability rather than its correctness.

Five domains were evaluated (landing pages, desktop applications, ad images, brand assets, product videos) across three phases of the creative workflow (ideation, mockup, refinement). Nearly 15,000 individual judgments were collected via pairwise comparisons aggregated by a Bradley-Terry model to produce ELO scores, Likert ratings on prompt adherence, usability, and visual appeal, and qualitative feedback.

No single model dominates all three phases in any domain. Claude Opus 4.6 leads in ideation, Gemini 3.1 Pro Preview takes over at the mockup stage when design system constraints come into play, GPT 5.3 Codex and Grok Imagine Video gain ground in refinement. Veo 3.1, strong in ideation, declines across each dimension as the task becomes more specific.

For ad images, visuals rated 5 out of 5 for usability ended up in the top 2 of pairwise comparisons in 84% of cases, compared to 10% for those rated 1, making usability the most reliable predictor of competitive success. The conclusion addressed to model developers: adherence to best practices and creative steerability are orthogonal axes that call for distinct optimization choices.