OpenAI announced on August 1, 2026, that its Astra model solved 10 long-standing problems in mathematics and theoretical computer science. The company stated that these areas had seen no progress for at least a decade, with total compute costs for the results reaching approximately $2,000.
Researchers, including Steven Miller from Yeshiva University and Francesco Fournier-Facio from the University of Cambridge, criticized the announcement. They argued that the model relied on prior work, including research published by Miller and Andreas Thom in 2016, which OpenAI failed to properly attribute.
Following the backlash, Scientific American published a report on August 6, 2026, labeling the company's actions as research misconduct and plagiarism. OpenAI subsequently revised some of its original statements regarding the model's achievements.
This incident follows previous controversies involving OpenAI's mathematical claims. In October 2025, a post regarding GPT-5 solving problems related to Paul Erdos’s conjectures was deleted. Conversely, a May 20, 2026, claim about disproving a 1946 conjecture received support from mathematicians Noga Alon and Thomas Bloom.
