Managed Model Inference Brings AI into Real-Time Data Streams
Managed Model Inference for LLMs and embedding models is now generally available in Confluent Cloud. Forecasting, anomaly detection, and text analysis models are also available in early access.
Also shipped Nov 18, 2026Confluent in November 2026 →
Genie Code CLI is now generally available
Databricks has made the Genie Code CLI generally available, aligning its management with workspace Genie Code. Administrators can oversee a single Genie Code budget across both the CLI and workspace, while users can apply free monthly usage in the CLI.
Smart Routing now supports external models for coding agents (Beta)
Databricks added support for routing coding-agent subagents to external model providers in the Unity Gateway Managed Configuration beta.
V3 progress control | CodeRabbit Cloud, V3 review settings preview
CodeRabbit updated progress controls in the V3 review settings preview, allowing progress_updates configuration to control review-progress reporting format.
Hyperdrive, Workers VPC - Custom CA and client certificates (mTLS) for Workers VPC
Cloudflare Workers VPC now supports custom certificate authority (CA) verification and mutual TLS (mTLS) authentication for private VPC Services.
Spark Monitoring UI (General availability)
Snowflake announced the general availability of the Spark Monitoring UI in Snowsight for tracking and troubleshooting Snowpark Connect for Spark jobs and sessions.
Amazon OpenSearch Memory is now generally available
Amazon OpenSearch Memory is now generally available as a serverless capability in Amazon OpenSearch Service to store, extract, and retrieve persistent memory for AI agents.
More from Confluent CloudEvery Confluent Cloud release →
What’s New in Q4’26 Confluent Data Streaming Platform: The Most Complete, Performant, and Open Foundation for Streamhouse
Confluent released updates including Lightning Tables, Managed Model Inference, and multi-tenancy for CPC. The release also adds Tableflow support for Google Cloud to improve streamhouse architectures.
Apache Flink 2.3 support for SQL statements in CMF 2.5.0
Compute pools in Confluent Manager for Apache Flink 2.5.0 can run statements on Flink 2.3 in addition to Flink 1.19. Flink 2.3 is required for process table functions, batch statements, model inference, and topic-subject associations.
Shared SQL compilers and compile workers in CMF 2.5.0
CMF compiles SQL statements for the Flink version of their compute pool using compilers running in extra containers within the CMF pod. Statements for environments with artifacts compile on dedicated compile worker pods.
Process table functions in CMF 2.5.0
Users can register user-defined process table functions that accept a table input and return a table. These functions can be called from statements running on Flink 2.3 compute pools.
Batch statements support in CMF 2.5.0
Statements can run as Flink batch jobs that process bounded input, write results, and finish. The runtime mode is reported in the statement status traits.
SQL views support in CMF 2.5.0
Users can create views in the environment catalog using CREATE VIEW and read from them directly inside SQL statements.
Model inference for OpenAI in CMF 2.5.0
Models for the OpenAI provider can be registered with CREATE MODEL and invoked with ML_PREDICT from statements on Flink 2.3 compute pools once endpoints are allowed by an administrator.
AI_FORECAST and AI_DETECT_ANOMALIES functions (preview)
New AI functions enable real-time time-series forecasting via AI_FORECAST and anomaly detection via AI_DETECT_ANOMALIES using pretrained foundation models deployed in your Kubernetes cluster.