Arena Flags GPT-6 Astra and Claude Fable 5.1 as Costly Leaders in Agent Benchmarking

Arena's latest agent-model comparison puts GPT-6 Astra and Claude Fable 5.1 at the expensive edge of frontier AI performance, highlighting a trade-off between measured task improvement and real-world cost. The ranking compares the top model from each family by net improvement score and median cost per task, with GPT-6 Astra and Claude Fable 5.1 standing out because both deliver strong gains while costing substantially more per task than other models from their respective labs.