inference Releases

This release enables email sending via workflows using Roboflow's email relay and introduces WebRTC data decoupling to separate video and data channels for bandwidth efficiency. It adds a Dinov3 classifier and makes GPU optional for CPU-only functions. Maintenance updates include workflow validation and improvements to OPC-UA integration.

Nov 17, 2025

The release focuses on stream processing capabilities by adding RTMP ingestion and processing with WebRTC streaming back, alongside Jetson 6.2.0 build support. It introduces a `USE_INFERENCE_EXP_MODELS` flag for model selection and enables TensorRT (TRT) compilation within the CLI. Maintenance updates include fixes for bursty RTSP streams and dependency management.

Nov 7, 2025

This release introduces hosted WebRTC support for real-time video streaming, adds RLE format support for SAM2, and includes various bug fixes for segmentation models and documentation.

Oct 31, 2025

Major update focusing on NPU support (Qualcomm QNN, MediaTek), CPU acceleration via SME2 instructions, and LLM optimizations including EAGLE-3 speculative decoding and HQQ quantization.

Oct 31, 2025

Updated the Workflow Execution Engine to v1.7.0 for improved error reporting, while adding new experimental models like PaliGemma 1 and YOLACT to the inference-experimental package.

Oct 24, 2025

Focused on bug fixes, this release addresses issues with RFDetr class remapping, multi-class keypoints detection, and API key authentication, alongside the addition of YOLOv8 classification support.

Oct 21, 2025

Addressed bugs in mask decoding for RF-DETR Seg and preprocessing blocks, while adding example scripts for GCS/S3 storage and improving benchmark reporting.

Oct 13, 2025

Introduced a critical fix for potential memory (OOM) errors caused by setting low confidence thresholds, adding a new environment variable to control the minimum allowed confidence.

Oct 3, 2025

Major release introducing RF-DETR Seg (Preview) for real-time instance segmentation, alongside performance improvements for mask processing utilities.

Oct 2, 2025

Minor release increasing the default resolution limit for RFDetr and adding a query parameter to dynamically disable model monitoring.

Oct 1, 2025

Added Claude Sonnet 4.5 to the Anthropic workflow block and resolved TensorRT (TRT) implementation issues for YOLOv8-pose.

Sep 30, 2025

Added EasyOCR support for text extraction and significantly improved OCR performance, alongside foundational support for Torch script backends in experimental features.

Sep 29, 2025

Enhanced the platform by adding statefulness to custom code blocks and fixing memory leaks in PolygonZoneBlock, alongside various remote execution and registry authentication updates.

Sep 19, 2025

A maintenance release focused on fixing authentication issues for Direct Deployment and Serverless legacy endpoints.

Sep 17, 2025

Performance improvements and authentication updates designed to speed up workflow block requests and secure serverless operations.

Sep 17, 2025

A major stability release featuring thread-safe execution for state-sensitive models and support for Modal sandbox execution.

Sep 13, 2025

Bug fixes addressing model lock acquisition errors and adding support for the SAM2-real-time model in the experimental package.

Sep 5, 2025

A quick fix release correcting a bug in multipart request parsing that was causing the request body to be read twice.

Aug 29, 2025

Architectural redesign of the inference server to eliminate event-loop blocking and introduction of a new dashboard for monitoring.

Aug 29, 2025

Significant update to the Execution Engine and addition of new visual blocks for workflows.

Aug 26, 2025