New Release: Ultralytics v8.4.57

Discussion
, ,
1

:rocket: Ultralytics v8.4.57 Released

We’re excited to share Ultralytics v8.4.57 — a release focused on Python 3.13 readiness, more reliable export workflows, and a meaningful RT-DETR CoreML export upgrade. It also includes a nice quality fix for semantic segmentation mask alignment and a broad set of documentation improvements.

If you’re building with Ultralytics YOLO, deploying to Apple devices, or preparing for newer Python environments, this release should make things smoother and more predictable. :raising_hands:

:glowing_star: Summary

v8.4.57 is primarily a stability and compatibility release rather than a new model release:

  • :snake: Python 3.13 is now a major CI focus
  • :outbox_tray: Export and slow-test CI are more robust and easier to debug
  • :red_apple: RT-DETR CoreML export is improved
  • :bullseye: Semantic segmentation masks are better aligned
  • :books: Docs were refreshed, including new guides for the Ultralytics Platform

:sparkles: New Features and Major Improvements

:snake: Python 3.13 readiness

This release makes a big step toward smoother support for Python 3.13:

  • CI testing now targets Python 3.13 across Linux, Windows, macOS, ARM, GPU, and benchmark jobs
  • Added official Python 3.13 package support metadata
  • Updated dependency constraints for export-related packages including TensorFlow, ONNX, CoreML, NumPy, and protobuf

This should help users moving to newer Python versions see more predictable install and test behavior.

:outbox_tray: More resilient export and CI workflows

A lot of work in this release improves reliability behind the scenes:

  • Added a dedicated slow-test runner with clearer diagnostics
  • CI now captures more helpful logs for crashes, memory issues, and dependency conflicts
  • Export tests have better isolation to reduce flaky failures from parallel execution

These changes are especially valuable for maintainers, contributors, and advanced deployment users. :magnifying_glass_tilted_left:

:shield: Clear compatibility guards for unsupported export stacks

Instead of failing mysteriously, unsupported combinations are now blocked or skipped more clearly:

  • TensorFlow export is blocked on macOS + Python 3.13
  • TensorFlow.js export is blocked on Python 3.13
  • Axelera export is skipped on Python 3.13
  • DEEPX CI validation remains on Python 3.12 for now due to upstream wheel availability

That means less trial-and-error and faster debugging. :white_check_mark:

:red_apple: RT-DETR CoreML export upgrade

One of the biggest model-related improvements in this release is better RT-DETR CoreML mlprogram export support:

  • RT-DETR-L exports more reliably to CoreML
  • Decoder operations were adjusted to reduce FP16 precision drift
  • New tests were added to validate the export path

This makes RT-DETR deployment on Apple hardware much more practical. :mobile_phone:

:bullseye: Better semantic segmentation mask alignment

A subtle but important fix improves segmentation output quality:

  • Resolved a mask offset issue caused by sub-pixel LetterBox padding
  • Masks are now resized more carefully before padding removal
  • Results should appear cleaner and better aligned on original images

Users working with segmentation may notice better edge placement and more accurate mask positioning. :framed_picture:

:books: Documentation Updates

This release also includes a strong docs refresh:

  • New Ultralytics Platform documentation integrations content for importing from Ultralytics HUB and Roboflow
  • DEEPX docs and branding updates
  • Fixes for broken links and inaccurate docs across Comet, TensorRT, Edge TPU, TF.js, ExecuTorch, OBB, and more
  • Explorer legacy snippets were updated to use YOLO11 models that match the pinned package version

For new projects, we continue to recommend YOLO26 as the latest stable Ultralytics model, with YOLO11 still fully supported for existing workflows.

:wrench: What Changed

:raising_hands: New Contributors

A big thank you to our new contributors:

:play_button: Try It Out

You can upgrade with:

pip install -U ultralytics

Then explore the latest release in the v8.4.57 release page or browse the full diff in the full changelog for v8.4.57.

:speech_balloon: Feedback

Please give v8.4.57 a try and let us know how it works in your environment — especially if you’re testing Python 3.13, export pipelines, RT-DETR on CoreML, or segmentation workflows. Your feedback helps the whole community improve faster. :rocket: