On the TerminalBench Science 0.1 benchmark for agentic scientific research, Fable 5.2 scored 6 points ahead of GPT-6 Astra