Databricks AI Index: token usage data from 10k+ enterprises shows cheaper tokens don't always mean cheaper agents
Original title: Databricks published token usage data from 10k+ enterprises. Cheaper tokens don't always mean cheaper agents
Databricks launched the AI Index, which aggregates anonymized usage data from 10,000+ enterprises via Unity Gateway plus a coding agent benchmark, updating weekly. The author highlights that Gemini 3.8 Flash's low token price still averaged $11.97 per task at 140 turns, while GPT-6 Luna and Claude Opus 5.5 matched 89.1% success at $0.05 vs $1.62 per task. Single-provider setups fell from 64% to 28%, open-weight models rose to 30.9% of token traffic, and the author advises measuring cost per completed task on your own workloads rather than leaderboards.
The post distills Databricks' task-level cost data, showing why cheap tokens don't mean cheap agents—a useful reference for model selection.
The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.