AI models' 'creative' output is becoming similar across providers

New research suggests that AI models from rival companies are becoming more alike in their creative answers, potentially narrowing the range of ideas that users encounter. New research from the US has found that ‘creative’ output across a wide range of leading AI models is becoming more similar over time. The authors, from Duke University, tested 69 models across 12 provider families, covering releases from 2023 to 2026, and found a statistically significant decline in output diversity across both real-world open-ended questions, and a standard creativity test – suggesting that similar ideas may be offered in response to ‘creative’ requests, increasingly, among all the major LLM providers. The provider families tested were Anthropic; Cohere; DeepSeek; Google; Meta; MiniMax; Mistral AI; Moonshot AI; OpenAI; Qwen; xAI; and Z.ai*: From the new paper – model releases used in the study, from March 2023 to July 2026. The chart covers 69 mode versions across 12 AI providers, demonstrating how the models tested were distributed across the three-year period, and showing increasingly frequent release schedules for some providers, which has to be accounted for in the authors’ reckonings. Source