AI news ·
OpenAI Updates GPT-6 Luna With Shorter Responses and Stronger Safety Scores
OpenAI made GPT-6 Luna respond more concisely and less agreeably, with better safety scores, but a health benchmark dropped, showing trade-offs.
OpenAI has updated GPT-6 Luna to produce shorter responses and to be less agreeable. The company also reported improvements across several safety scores.
The update did not lift performance uniformly. One health-related benchmark declined. This trade-off underscores the need for multidimensional evaluation when modifying models, because gains in one area can hide regressions elsewhere.