OpenAI evaluates model performance on personal admin benchmarks

OpenAI created a test benchmark named LifeVal to measure how well AI assistants perform everyday personal chores, including booking appointments, filing legal forms, and managing bills. The company reported that its GPT-6 Astra model achieved an 80.6 percent success score, though critics point out the benchmark was evaluated internally without independent oversight.