Claude Sonnet 5.5 Comes to Databricks: Unified Governance Across Three Cloud Platforms

Claude Sonnet 5.5 Joins the Databricks Ecosystem

Anthropic's Claude Sonnet 5.5 model is now officially available on the Databricks platform, spanning AWS, Azure, and GCP — with unified governance through Unity Gateway. This means enterprise users, regardless of which cloud they're deployed on, can access the model consistently while keeping it under a single permissions and compliance management framework.

For enterprises that have already built their data assets on Databricks, having the model and data on the same platform is significant. It reduces the security risks and engineering complexity of moving data across platforms, and allows model calls to plug directly into existing data governance frameworks.

Performance Positioning: Efficiency Gains and Stronger Document Understanding

According to official descriptions, Claude Sonnet 5.5 improves on its predecessor Sonnet 5 in coding tasks and agentic use cases. These are precisely the two areas at the heart of enterprise AI adoption today — code generation accelerates development, while agents handle multi-step automated workflows.

Perhaps more noteworthy: Sonnet 5.5 achieves Opus 5-level accuracy in document understanding, parsing, and search. This is a meaningful signal. Tasks that previously required the more capable (and more expensive) Opus-series models can now be handled by the lighter-weight Sonnet series. For enterprises dealing with high volumes of unstructured documents — contracts, reports, knowledge bases — this means maintaining accuracy while potentially reducing inference costs.

A Rich Model Portfolio and Toolchain

Sonnet 5.5 doesn't arrive in isolation. It joins an existing lineup on Databricks that includes Claude Opus 5.5, Claude Fable 5.1, and a library of 60+ open-source and frontier models. This multi-model strategy lets enterprises match the right model to each task — from high-precision flagship models to cost-efficient lightweight ones — creating a flexible, tiered approach to model usage.

Around these models, Databricks offers a complete toolchain:

Agent Bricks: Building Domain-Specific Agents

Developers can use Agent Bricks to build agents tailored to specific business domains, focusing and constraining the capabilities of general-purpose LLMs to concrete industry and task contexts.

Agent Bricks is Databricks' agent-building framework. Its core approach is to "specialize" general-purpose LLMs through tool binding, retrieval-augmented generation (RAG), and system prompt constraints — focusing them on specific industries or tasks. Compared to calling raw model APIs directly, Agent Bricks encapsulates agent orchestration logic, allowing developers to declaratively define an agent's knowledge boundaries, callable external tools (such as database queries or API calls), and execution strategies — without building a scheduling framework from scratch. This lowers the engineering barrier for embedding LLM capabilities into internal business processes, and allows different business units to maintain their own specialized agents on the same platform while sharing underlying model resources and governance mechanisms.

Databricks Apps and Lakebase Memory

Completed agents can be deployed as Databricks Apps, and with the memory capabilities provided by Lakebase, agents can maintain context and state across multi-turn interactions. This fills a critical gap in moving agent applications from "one-off Q&A" to "ongoing collaboration."

Unity Gateway: End-to-End Governance

All model calls are governed through Unity Gateway. In enterprise AI deployments, governance is often the invisible threshold that determines whether scaling is even possible — who called the model, what they called, whether it was compliant, how costs are allocated. These all require a unified control point. Unity Gateway plays exactly that role.

Unity Gateway is built on Databricks' Unity Catalog data governance system, which has long handled permissions management, lineage tracking, and compliance auditing for data assets. Bringing model calls under the same governance framework means enterprises can use the same policies and tools they use to manage data assets to manage AI inference behavior — for example, setting model access permissions by department, logging inputs and outputs for each call to satisfy audit requirements, and tracking inference costs by project. This is especially important for heavily regulated industries: financial institutions need to demonstrate the traceability of AI decisions, healthcare organizations need to ensure patient data isn't sent to unauthorized models, and Unity Gateway's unified control layer can directly address these compliance requirements without building a separate audit infrastructure for the AI layer.

What This Means for Enterprise AI Adoption

Bringing Claude Sonnet 5.5 into Databricks is fundamentally about further merging "model capability" with "data platform + governance." For enterprise decision-makers, choosing a model is no longer just about point-in-time performance — it's about whether the model can integrate into existing data and governance systems, how smooth deployment is, and whether costs are manageable.

Sonnet 5.5's document understanding performance approaching Opus 5, combined with Databricks' tools for agent building, deployment, and governance, offers document-intensive industries — such as finance, legal, and healthcare — a more cost-effective path to AI adoption. As the model library continues to expand, enterprises' ability to "choose the right model for the right job" on a single platform will only grow stronger.

For more technical details, refer to the official Databricks documentation.