GMI Cloud on Artificial Analysis Intelligence Index v5: Private Test Data and Longer Output Runs

Key Info

GMI Cloud shared two observations on the newly announced Artificial Analysis Intelligence Index v5: 60% of the score now comes from private test data, and the median frontier model now uses 59k output tokens per task, up from 35k.

Highlights

• Artificial Analysis Intelligence Index v5 adds 2 new benchmarks and 3 upgrades, with the majority of the Index built on private datasets for the first time.
• 60% of the score now comes from private test data, making it harder for models to train on the test data.
• The median frontier model now consumes 59k output tokens per task, up from 35k.
• GMI Cloud argues this shift means infrastructure needs predictable throughput and cost to support longer, multi-step runs.