OpenAI quietly updates its evaluation metrics for GPT-6 Astra, making changes that appear to favor Astra and continuing to revise other metrics after launch (Emily Forlini/Fortune)

OpenAI revised multiple evaluation benchmarks for its GPT-6 Astra model, adjusting metrics in ways that advantage Astra’s performance scores. The company also announced a new framework to log misalignment incidents throughout training, evaluation, and deployment, citing the recent “wiki incident” as a catalyst.