Version 11.6.0¶

March 7, 2026

This page contains the new features, enhancements, and fixed issues for DataRobot's Self-Managed AI Platform 11.6.0 release. これは長期サポート（LTS）リリースではありません。 リリース11.1が最新の長期サポートリリースです。

Version 11.6.0 includes the following new features and fixed issues.

エージェント型AI¶

New and retired LLMs¶

With this release, OpenAI GPT-5.2 is available through the LLM gateway. 従来通り、組織の特定のニーズに対応するために外部連携を追加できます。 サポートされているLLMの完全なリストについては、利用可能なLLMのページを参照してください。

In addition, the following LLMs are retired:

• GPT-4o Mini (retired February 27, 2026)
• Cerebras Qwen 3 32B (retired February 16, 2026)
• Cerebras Llama 3.3 70B (retired February 16, 2026)
• Mistral (7B) Instruct v0.2 (retired February 25, 2026)
• Marin Community Marin 8B Instruct (retired February 25, 2026)

Agent Assist¶

The 11.6.0 release introduces Agent Assist (dr-assist), an interactive AI assistant optimized for the development of AI agents. It helps users design, code, and deploy agents through natural conversation—users describe the agent they want, and the assistant helps build it on the foundation provided by the Agentic Starter application template.

Agent Assist integrates with the DataRobot CLI as a plugin and uses the DataRobot LLM gateway for model access. During the design and code cycle, Agent Assist can outline which tools an agent should call based on the proposed functionality—for straightforward tools, it can implement the tool code; for more complex tools (such as those that consume API tokens or write to a database), it can scaffold the initial file structure for the human-in-the-loop to complete in the editor or development environment of their choice.

Agent Assist can:

• Design AI agents by helping users think through specifications, ask clarifying questions, and produce an agent specification file (agent_spec.md).
• Research solutions using file search and analysis (an internal agent can read files, list directories, grep, and glob).
• Code AI agents by loading an existing agent_spec.md, cloning the DataRobot agent template repository, and implementing the agent with file edits and shell commands.
• Simulate an agent from a specification before coding—rehearsal mode lets users try the design interactively to verify the functionality outlined by the specification.
• Deploy agents to DataRobot following the template’s deployment instructions.

予測AI¶

Incremental learning now supports dynamic datasets¶

増分学習（IL）は、10GB〜100GBのデータセットを利用する教師ありエクスペリメントに特化したモデルトレーニング方法です。 データをチャンク化し、トレーニングのイテレーションを作成することで、予測を行うための最適なモデルを特定できます。 This release enables support for using incremental learning on dynamic datasets of any size, whereas static datasets must be between 10GB and 100GB.

データ¶

Database connectivity UI now uniform across NextGen¶

This release implements a standardized user interface when working with data connections across NextGen, providing a more unified experience. This update includes the following areas:

• In Registry > Data > Add data.
• The Browse data modal in Workbench.
• In Account settings > Data connections.
• The vector database creation workflow.

Previously, the interface may have been significantly different based on where you accessed database connectivity.

Support for Trino connector added to DataRobot¶

Support for the Trino native connector has been added to DataRobot, allowing you to:

• データ接続を作成して設定する。
• Upload data from Trino into DataRobot.
• Use as an intake source and output destination for batch prediction jobs.

MLOps and predictions¶

Asset lineage graph view¶

The Lineage view provides visibility into the assets and relationships associated with a given MLOps artifact. This view is available on a deployment’s overview tab in Console, in the version details for a registered model in Registry, and in the version details for a custom model in Workshop. In each location, the Lineage section helps you understand the full context of the asset—including models, datasets, experiments, deployments, and other connected artifacts—so you can review AI systems and track how assets relate.

The Graph tab shows an interactive, end-to-end visualization of those relationships as a DAG (Directed Acyclic Graph) made up of nodes (assets) and edges (relationships). The asset you are viewing is highlighted with a purple outline. When reviewing edges, solid lines represent concrete, persistent relationships within the platform, such as a registered model used to create a deployment. Dashed lines Arrows generally flow from the "ancestor" or container to the "descendant" or content (for example, from a registered model version to a deployment).

For details on the graph and the available controls, see the deployment overview, registered models, and custom model versions documentation.

Scaled to zero prediction service improvement¶

This release increases the chat completion prediction service wait timeout to improve reliability for agentic workflow and custom model deployments using "scale to zero" optimization. When a deployment scaled to zero receives its first prediction request, a new server must be provisioned. The previous 20-second wait timeout was often too short for a new server to become ready, resulting in a "bad gateway" response. This update increases the prediction service wait timeout from 20 seconds to 300 seconds (5 minutes), mitigating the occurrence of "bad gateway" responses when the initial server provisioning takes longer than 20 seconds.

プラットフォーム¶

View resource usage information for your account¶

All users can now view resource usage information in account settings, providing greater visibility into graphics processing unit (GPU), central processing unit (CPU), and large language model (LLM) API usage across the platform. To access usage information, open Account settings > Usage Explorer. From this page, you can view resource consumption by service for a given date range, as well as export the report as a CSV file. Administrators can access an additional dashboard from Admin settings > Tenant Usage Explorer (previously named “Usage Explorer”).

管理¶

非ビルダーユーザーを組織に追加する¶

管理者が、組織に非ビルダーユーザーを追加できるようになりました。 Non-builder users can only access the applications associated with the organization they’re joining, as well as basic user settings. アプリケーションを操作する際、予測の実行、プロンプトの追加、チャットの開始、表示/削除、およびデータのアップロードを行うことができます。

非ビルダーユーザーを追加するには、既存のユーザーアカウントに非ビルダーのシートライセンスを割り当てるか、新しいユーザーを招待する機能を使用して、一度に最大20名まで招待することができます。 なお、非ビルダーユーザーは、組織の最大アクティブユーザー割り当て数には含まれません。

OAuth for Google Drive support¶

This release streamlines DataRobot's OAuth connection process to services like Google Drive and Confluence by introducing a centralized, self-service OAuth system. This means you only have to set up and authorize your external account once, managing all your secure connections in a single spot. DataRobot then automatically retrieves temporary access tokens when needed to ingest your data. This standardization makes connecting easier and more secure, and it will enable these connectors to be used in more DataRobot areas like Apps and Model Creation Projects. For information on how to configure OAuth connection for the providers supported, see OAuth provider management.

コードファースト¶

Python client v3.13¶

Python client v3.13 is now generally available. For a complete list of changes introduced in v3.13, see the Python client changelog.

DataRobot REST API v2.42¶

DataRobot's v2.42 for the REST API is now generally available. For a complete list of changes introduced in v2.42, see the REST API changelog.

Issues fixed in Release 11.6.0¶

Applications fixes¶

• APP-5390: Adds the ability for app builders to enable usage of their assets by other users inside the app, granting non-builders access to data stores and registry items so that they can find value in the app.

• APP-5462: Fixes an issue where unauthenticated access to an application did not redirect back to the application after login when that login used SSO.

• APP-5463: Non-builder users with only one application are now automatically redirected from the Application Gallery to their application.

データの修正¶

• DM-20379: Fixes an issue with Azure OAuth that was preventing token acquisition.

コアAIの修正¶

• MMM-21322: For deployments that are using agentic workflow models, checks for specific cases when a deployment doesn't provide association_id or PROMPT_COLUMN_NAME and helps the user navigate to the issue so that they can see the tracing logic.

• MODEL-22260: Resolves an issue where the tileserver-watcher container failed to validate TLS when connecting to object storage with a private CA TLS issuer.

• MODEL-22621: Returns early when there are no projects associated with the user, resolving HTTP 500 / BSONObjectTooLarge in dr.Project.list.

• PRED-12306: The default date range for the deployment quota usage monitoring page is now limited to 90 days to prevent errors from the OTel Collector API when requesting longer ranges.

• RAPTOR-15577: Fixes the decision logic for blocking the Deploy button in Registry so that it is aligned with the correct calculation of limits during the deployment creation phase. Now, the Deploy button is blocked if the custom model deployment usage quota is reached.

プラットフォームの修正¶

• CMPT-4664: Disables unnecessary "missing tenant context" logs during internal health checks to prevent k8s_workers_report from filling logs.

• CMPT-4667: In scanner mode, IBS now authenticates before the build.

• CMPT-4684: Fixes image deletion errors. For releases 11.1 to 11.4, installation images could not be deleted. Image deletion is now enabled, but old image metadata can be missing or corrupted, causing the Delete endpoint to fail. Now, a K8s job fetches and updates the image metadata from the registry, allowing the deletion endpoint to complete.

• CMPT-4705: Improves failure detection for builds when image scanning is enabled; init container errors now fail builds immediately instead of leaving them in ERROR and timing out.

• CMPT-4706: Replaces the build-id in the scan report with the image-uri so that if the scanner upload report is written to a previous location, it doesn't get overwritten.

• CMPT-4807: PostgresConnectivityHealthCheck now supports Postgres running on a non-standard port via EngConfig['PGSQL_PORT'].

• PLT-20382: Fixes an issue affecting the permanent deletion process. Previously, in certain scenarios, when an organization administrator attempted to permanently delete a user who had shared resources, DataRobot failed to complete the deletion.

• PLT-20525: Fixes an issue where seat license administration pages displayed incomplete data for organizations with more than 100 users.

• PLT-20599: The user provisioning process for LDAP/SAML now assigns a non-builder seat to new user account, if the respective subscription is applicable.

Notebooks fixes¶

• CFX-4797: Fixes an issue where JavaScript (.js) files did not open in the built-in Codespace editor.

• CFX-5124: Fixes an issue where users with OWNER permissions on a Use Case could not delete Notebooks or Codespaces shared with them via that Use Case.

記載されている製品名および会社名は、各社の商標または登録商標です。 製品名または会社名の使用は、それらとの提携やそれらによる推奨を意味するものではありません。