DeepSeek V4.1 Flash beats Astra on AA's new benchmark

Artificial Analysis (AA) recently updated its Intelligence Index to v4.3, introducing a new private evaluation that replaced τ3. Initially, Astra performed well, matching Fable. However, DeepSeek V4.1 Flash has now surpassed Astra, taking the top spot on AA's new benchmark. This occurred after AA reportedly adjusted the index twice in three days, seemingly to prevent Astra from appearing inferior to Fable.

This report is the first to highlight DeepSeek V4.1 Flash's top performance on AA's new benchmark, unlike earlier reports focusing on Astra's initial lead.