Enterprises are spreading AI work across more model providers, and the cheapest token is often not the cheapest task. Databricks launched its AI Index on Nov. 16, built on aggregated, anonymized data from more than 10,000 companies that reach models through its Unity Gateway. In July about 64% of them relied on one model provider. By Nov. 2 that was 28%. Open-weight models went from 7.3% of enterprise token traffic in July to 30.9% in November.
Share still moves quickly toward a strong release. Databricks says Claude Opus 5.5 went from 7.2% to 37.9% of proprietary-model tokens in six weeks, lifting Anthropic’s share from 57.6% to 68.8%. The index also prices work by the task. In its coding benchmark GPT-6 Luna and Claude Opus 5.5 both completed 89.1% of tasks, at $0.05 and $1.62 per task. Gemini 3.8 Flash has a low token price, yet it averaged 140 turns and cost $11.97 per task.
A second dataset points the same way on concentration. The Ramp AI Index, drawn from the spending of more than 85,000 U.S. businesses, puts Anthropic at 49.3% of AI token spending in the week ending Nov. 1. The two measure different things, tokens through one gateway and dollars through one card platform, and neither is a census. Both show a wide gap between the typical buyer and the heaviest one. Databricks says the median user spent $2.49 and the 99th-percentile user $699.74. Ramp’s median paying firm spends $16.32 per employee per month, and its top 1% spends $10,199.20.
In our reading, a commitment sized on mid-year usage with one provider deserves a second look, and budgets set on average spend per user will miss where the money goes. We covered the fall in Ramp’s blended token price and how AWS and Databricks meter coding agents by user and team. Databricks sells the gateway its data comes from, estimates spend from list prices and runs the benchmark on its own codebase.
Source: Databricks Blog