New kind of AI uses a fresh approach to reasoning —‬ researchers say it costs up to 11 times less to run than a leading OpenAI model

Live Science · 2026-08-29

Scientists at AI company Pathway have developed a new "post-transformer" AI model, BDH-CQ, which employs a vector-based approach to cognition that significantly reduces operational costs compared to traditional transformer-based models. The researchers suggest this marks the beginning of a "post-transformer" era in AI. Published on arXiv on August 10, the new model builds on a precursor, "Dragon Hatchling," created in 2025. BDH-CQ scored nearly 30% on the 2019 ARC-AGI-1 benchmark, a test measuring nonverbal reasoning through puzzles. Although other models have higher scores, BDH-CQ's architecture makes it dramatically smaller and cheaper to run. For instance, while OpenAI's GPT 5.6 Luna (Low) achieved a slightly higher score, it cost approximately 11 times more in token costs. BDH-CQ was trained on just 150 million parameters, and scientists believe its cognitive capabilities could scale significantly with larger parameter sizes, potentially impacting the cost and scale of AI deployments.

Read the original report at Live Science