feat(databricks): add system.ai.* model service entries

[Archit-Kumar-16]

Note

Low Risk
Additive metadata-only YAML with no runtime or auth changes; main risk is incorrect pricing or limits affecting cost estimates and routing for new model IDs.

Overview
Adds 24 new Databricks provider YAML definitions for system.ai.* foundation-model serving IDs, alongside existing databricks-* entries.

Coverage spans Claude (Haiku/Sonnet/Opus 4.x, including retired claude-sonnet-4), Gemini 2.5 and Gemma 3, GPT-5 family (including gpt-5-4 / gpt-5-5-pro tiered pricing and responses-only codex/pro variants), gpt-oss open models, Llama/Meta instruct models, and gte-large-en embeddings.

Each file records serverless provisioning, token costs (often with prompt-cache fields), limits/modalities, features, and per-model params, removeParams, supportedModes (chat vs responses), and lifecycle fields where relevant (retirementDate, deprecated claude-sonnet-4).

^{Reviewed by Cursor Bugbot for commit 04f2897. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 04f2897. Configure here.}

[cursor]

Gemini models wrongly marked active

Medium Severity

system.ai.gemini-2-5-flash and system.ai.gemini-2-5-pro set status to active, but Databricks already marks Gemini 2.5 Flash and Pro as deprecated with retirement on 2026-10-02. Callers that trust status for model selection will treat retiring endpoints as current.

Additional Locations (1)

• providers/databricks/system.ai.gemini-2-5-pro.yaml#L28-L29

 

^{Reviewed by Cursor Bugbot for commit 04f2897. Configure here.}

[cursor]

Claude Sonnet 4 wrongly retired

Medium Severity

system.ai.claude-sonnet-4 sets status to retired with retirementDate 2026-06-15, but on Databricks Claude Sonnet 4 is still deprecated and scheduled to retire on 2026-10-09. Consumers that exclude retired models will incorrectly drop a still-available Databricks endpoint.

 

^{Reviewed by Cursor Bugbot for commit 04f2897. Configure here.}