Databricks opens an enterprise AI index: equal task success, but a roughly 32x gap in cost per task
Synopsis
Databricks has launched a public AI index that combines anonymized usage data from more than 10,000 enterprises on Unity Gateway with coding-agent evaluations built on real tasks from its private codebase, updated weekly. The index shows enterprise model choice fragmenting quickly, with open-weight models rising from 7.3% of enterprise token traffic in July to 30.9% in November. In the coding benchmark, GPT-6 Luna and Claude Opus 5.5 both reached an 89.1% task success rate, yet averaged $0.05 and $1.62 per task respectively.
Interpretation
Enterprises are moving from concentration on a few providers toward spreading workloads across model versions and providers, while model quality still drives share shifts. In July about 64% of enterprises using Unity Gateway relied on a single model provider; by November 2 that had fallen to 28%. Within six weeks of launch, Claude Opus 5.5 rose from 7.2% to 37.9% of proprietary-model tokens, lifting Anthropic's overall share from 57.6% to 68.8%.
The index places two data sources side by side: actual enterprise usage and spending, and continuously updated coding-agent evaluations. Customers repeatedly asked how new models compare and what enterprise adoption looks like, and the index makes these internal evaluations and usage observations public together. Usage data comes from more than 10,000 companies on Unity Gateway, and evaluation tasks are drawn from real pull requests in Databricks' private codebase and scored by hidden tests rather than an LLM judge.
Coding-agent cost depends on token price together with the work a model needs to finish a task, so unit price alone misleads. Gemini 3.8 Flash has a low token price but averaged 140 turns per task and ended as one of the benchmark's most expensive models at $11.97 per task. In the same benchmark, Claude Fable 5.1 reached a 96.0% task success rate at $7.47 per task while Muse Wonder reached 94.9% at $1.53, so 1.1 extra points came at nearly five times the cost.
Perspective
The index is aimed at enterprise readers comparing model adoption, coding-agent performance and cost; the dataset covers September 21 through November 7, 2026, and spending is estimated from publicly listed provider prices. Results reflect the workloads and models evaluated and may vary across organizations.
The index states that results may vary across organizations and recommends that enterprises evaluate models against their own development tasks; three open-weight models rank in the top 10 for coding quality, with DeepSeek V4.1 and GLM 5.3 tied for third at a 93.1% pass rate.
