Benchmarking Coding Agents on Databricks’ Multi-Million Line Codebase
Summary
Internal benchmarking across Databricks’ multi-million line codebase revealed that per-token pricing is a misleading indicator of actual development costs compared to model reasoning efficiency and harness context management. The evaluation shows teams can significantly reduce expenses without sacrificing quality by matching task complexity to tiered models and adopting cost-effective daily drivers like GLM 5.2 for routine engineering work.
Summary generated by brickster.ai. For the full article, follow the source link above.
Topics
More from Databricks Blog
The lakehouse is a better data warehouse: 2026 benchmarks and proof
Recent benchmarks show Databricks Lakehouse performance improved 77% since 2022, driving 25-60% cost reductions and 80-90% faster analytics for migrating organizations. Beyond matching core data warehouse capabilities, the lakehouse unifies AI, streaming, unstructured data, and open-format portability on a single governed platform.
Scaling and Operating a Large dbt Project on Databricks: IFCO's Data Team on Performance, Visibility, and Debugging
Tuning dbt incremental models with liquid clustering, dynamic file pruning, and deliberate merge strategies cut IFCO's core job runtime by over 60% and retired their nightly full refresh. By diagnosing real executed query plans and orchestrating models as discrete Databricks Jobs tasks via the open-source databricks-dbt-factory, the team gained per-model visibility, targeted reruns, and enforced testing.
Meta’s ads MCP server comes to Databricks: Put your customer intelligence to work in advertising campaigns
Meta’s ads MCP server is now available in the Databricks Marketplace, allowing marketers to use natural language in Genie to evaluate campaign performance alongside governed business data. Administrators can control access to the connection through Unity Catalog, with Unity Gateway governing tool calls and recording audit logs.
NEAREST BY Join: Scaling Vector Search in Databricks Runtime
Databricks Runtime now features NEAREST BY, a new SQL join that executes exact or approximate batch vector searches directly against your Lakehouse data. Powered by a fused Photon operator with a custom blocked GEMM kernel, it turns standard liquid-clustered Delta tables into partition-pruned vector indexes with no external vector store to sync or manage.
Unlocking Data Portability: Preventing Catalog Lock-in with REGISTER and UNREGISTER APIs
The new REGISTER and UNREGISTER APIs provide a standard method to safely transfer open table format management between catalogs without copying data stored in customer-owned buckets. By requiring the original catalog to explicitly relinquish control before a transfer, this workflow prevents catalog lock-in and eliminates dangerous split-brain scenarios.
How Genie Ontology powers product development at Databricks
Databricks relies on Genie Ontology and Genie One to drive product development, combining human-curated semantics and learned knowledge across first- and third-party sources to analyze adoption and forecast trends. The platform extends these internal workflows with capabilities including custom skills, scheduled tasks, shareable agents, and MCP writes to external tools.