v0.62.4

roboflow/inferencev0.62.4Dec 10, 2025by grzegorz-roboflow

AI Summary

The update significantly expands the `inference-exp` module with new OCR models like DocTR, EasyOCR, and TROcr, alongside GroundingDino support. It introduces new versions of Claude and OpenAI workflow blocks and adds Roboflow Managed API Key support for Google Vision OCR. The release also includes a complete WebRTC SDK implementation and various maintenance fixes for Modal and builds.

Key Highlights

  • DocTR, EasyOCR, and TROcr in `inference-exp`
  • Claude workflow block v2
  • OpenAI workflow block v4
  • Roboflow Managed API Key for Google Vision OCR
  • Full WebRTC SDK implementation

New Features

  • DocTR
  • EasyOCR
  • TROcr
  • GroundingDino
  • Claude v2
  • OpenAI v4
  • WebRTC SDK
  • Roboflow Managed API Key for Google Vision

Full Release Notes

# 💪 Added
* Feature/inference exp jp 5 by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1728
* Add tests and fixes for DocTR model by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1777
* E2E platform tests for `inference-exp` to run in CI by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1774
* Claude workflow block v2 by @brunopicinin in https://github.com/roboflow/inference/pull/1757
* Add notice about custom python support in batch processing by @SolomonLake in https://github.com/roboflow/inference/pull/1754
* Add EasyOCR to `inference-exp` by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1780
* Add Roboflow Managed API Key support to Google Vision OCR by @brunopicinin in https://github.com/roboflow/inference/pull/1767
* OpenAI workflow block v4 by @brunopicinin in https://github.com/roboflow/inference/pull/1762
* Add TROcr to `inference-exp` by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1781
* Adds NMS deduplication to SAM3 by @lrosemberg in https://github.com/roboflow/inference/pull/1776
* Add support for gaze model in `inference-exp` by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1783
* `GroundingDino` registered in `inference-exp` by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1786
* New upload format for webrtc file by @digaobarbosa in https://github.com/roboflow/inference/pull/1778
* WebRTC - python sdk by @digaobarbosa in https://github.com/roboflow/inference/pull/1701
* Turn server sdk config by @digaobarbosa in https://github.com/roboflow/inference/pull/1715

# 🔧 Fixed
* when runnning on modal we get 500 for workspace id params by @digaobarbosa in https://github.com/roboflow/inference/pull/1759
* Address 'RuntimeError: lock is bound to a different event loop' in modal by @grzegorz-roboflow in https://github.com/roboflow/inference/pull/1750
* Fix/depth anythig v2 fixes by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1771
* Try to fix builds broken by `rf-mediapipe` dependency by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1785
* Fix first modal startup timeout calculation; add quota check by @grzegorz-roboflow in https://github.com/roboflow/inference/pull/1782
* Add missing 'engine_host_code_allowed' flag for YOLOv10 implementation in TRT by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1791
* When encapsulating webrtc worker within exception block, do not return in finally by @grzegorz-roboflow in https://github.com/roboflow/inference/pull/1792

# 🏡 Maintenance
* Add connection pooling and retry logic to OPC UA writer by @alexnorell in https://github.com/roboflow/inference/pull/1751
* Add examples showing how to use SAM3 hot instances by @grzegorz-roboflow in https://github.com/roboflow/inference/pull/1756
* Specify public stun servers by @grzegorz-roboflow in https://github.com/roboflow/inference/pull/1766
* Integration tests inference sdk by @digaobarbosa in https://github.com/roboflow/inference/pull/1702
* Parametrize modal webrtc worker watchdog timeout by @grzegorz-roboflow in https://github.com/roboflow/inference/pull/1769
* Tweaks to improve data channel frames processing by @grzegorz-roboflow in https://github.com/roboflow/inference/pull/1772
* Usage collector model_id by @grzegorz-roboflow in https://github.com/roboflow/inference/pull/1768
* Remove grounding dino extras by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1788
* Adjust terminal condition in modal to check if heartbeat ever occured rather than to check for heartbeat count by @grzegorz-roboflow in https://github.com/roboflow/inference/pull/1789
* Get rid of verification of named outputs in case of YOLO-NAS TRT models - as the artefacts are clearly inconsistent by @PawelPeczek-Roboflow in https://github.com/roboflow/inference/pull/1790
* OWLv2 models are logging each inference at INFO level, changing to DEBUG by @grzegorz-roboflow in https://github.com/roboflow/inference/pull/1793

**Full Changelog**: https://github.com/roboflow/inference/compare/v0.62.2...v0.62.4