How Databricks rolls out frontier models to 12,000 employees on Day 1

Databricks uses Unity Gateway to give all 12,000 employees day-one access to new models via an experimental label, then controls costs with four per-user budget types. The process relies on private benchmarks, user feedback, and OpenTelemetry cost tracking to decide whether to promote or retire each model within days. In a recent release week, Opus 5, GPT-6 Sol, and GPT-Luna were confirmed on the efficiency frontier within three days.

for who

Engineering platform and AI infrastructure teams managing model rollouts for large workforces.

why now

With new frontier models launching weekly, Databricks' Day-1 rollout method shows how to adopt cost-effectively.

what changes

They can now offer immediate, day-one access to new models for thousands of users while keeping cost overruns in check and promoting only models that prove their quality on internal benchmarks.

to do

Adopt a Gateway like Unity Gateway to distribute experimental model configs, enforce four per-user budget layers, and evaluate models with benchmarks, user feedback, and OpenTelemetry cost data before promotion.