# New Release: Ultralytics v8.4.203

**URL:** <https://community.ultralytics.com/t/new-release-ultralytics-v8-4-203/2299>\
**Category:** Discussion\
**Tags:** ultralytics-official, releases, announcements\
**Created:** [November 6, 2026, 1:05pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-203/2299 "2026-11-06T13:05:38Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![glenn-jocher](https://sea1.discourse-cdn.com/flex001/user_avatar/community.ultralytics.com/glenn-jocher/32/83_2.png) [@glenn-jocher](https://community.ultralytics.com/u/glenn-jocher)\
**Post date:** [November 6, 2026, 1:05pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-203/2299/1 "2026-11-06T13:05:38Z")

</div>

# 🚀 Ultralytics v8.4.203: NXP Export, Safer Loading, and Reliability Fixes

**Ultralytics YOLO v8.4.203 is here!** This release introduces **NXP i.MX95 Neutron NPU export** , strengthens dataset and checkpoint loading security, and improves training, inference, tracking, and visualization.

## 🌟 New Feature: NXP i.MX95 Export

[PR #27047](https://github.com/ultralytics/ultralytics/pull/27047) by [@glenn-jocher](https://github.com/glenn-jocher) adds `format=nxp` export for **all seven YOLO tasks**.

- Creates an **INT8 LiteRT model** compiled with **Neutron SDK 3.3.0**.
- Uses the existing LiteRT inference backend with NXP’s delegate.
- Targets **i.MX95 only** ; unsupported operations may fall back to the CPU.

Before deployment, validate accuracy, latency, and CPU fallback behavior on your target board.

## 🔒 Security Improvements

- **Restricted dataset cache loading:** [PR #27041](https://github.com/ultralytics/ultralytics/pull/27041) by [@onuralpszr](https://github.com/onuralpszr) introduces a restricted unpickler for dataset caches.
- **No automatic checkpoint dependency installation:** [PR #27035](https://github.com/ultralytics/ultralytics/pull/27035) by [@onuralpszr](https://github.com/onuralpszr) makes missing checkpoint dependencies raise an error instead of automatically installing packages.
- **Safer SAM and NAS loading:** [PR #27032](https://github.com/ultralytics/ultralytics/pull/27032) by [@Y-T-G](https://github.com/Y-T-G) uses weights-only checkpoint loading where supported.
- **Safer text embedding caches:** [PR #27045](https://github.com/ultralytics/ultralytics/pull/27045) by [@onuralpszr](https://github.com/onuralpszr) applies weights-only loading to YOLO-World and YOLOE text embedding caches.
- **Shell-free sudo detection:** [PR #27034](https://github.com/ultralytics/ultralytics/pull/27034) by [@onuralpszr](https://github.com/onuralpszr) replaces a shell call with `shutil.which`.

## 🧠 Training Improvements

- **YOLOE prompt-loss fixes:** [PR #27048](https://github.com/ultralytics/ultralytics/pull/27048) by [@raimbekovm](https://github.com/raimbekovm) corrects text-prompt training with smaller class counts and fixes CPU runner image builds.
- **Correct hand-keypoint flips:** [PR #27024](https://github.com/ultralytics/ultralytics/pull/27024) by [@raimbekovm](https://github.com/raimbekovm) preserves landmark assignments when hands are flipped.
- **Faster deterministic semantic loss:** [PR #27027](https://github.com/ultralytics/ultralytics/pull/27027) by [@Y-T-G](https://github.com/Y-T-G) reports a **2.8x speedup in focused measurements** , not a guaranteed end-to-end training speedup.
- **Cleaner Platform training commands:** [PR #27004](https://github.com/ultralytics/ultralytics/pull/27004) by [@laodouya](https://github.com/laodouya) removes duplicate training-command serialization.

## 🎯 Inference, Tracking, and Export Fixes

- **Better tracker assignments:** [PR #27026](https://github.com/ultralytics/ultralytics/pull/27026) by [@Y-T-G](https://github.com/Y-T-G) improves matching in the SciPy fallback and skips unused LAP cost computation.
- **Predictable FastSAM box prompts:** [PR #27043](https://github.com/ultralytics/ultralytics/pull/27043) by [@raimbekovm](https://github.com/raimbekovm) selects no masks when a box overlaps none.
- **Resilient SAM prompts:** [PR #27046](https://github.com/ultralytics/ultralytics/pull/27046) by [@raimbekovm](https://github.com/raimbekovm) keeps prompts usable after an interrupted “segment everything” call.
- **Consistent backend precision:** [PR #27040](https://github.com/ultralytics/ultralytics/pull/27040) by [@aswanth-07](https://github.com/aswanth-07) synchronizes precision after module conversions.
- **More reliable ExecuTorch exports:** [PR #27044](https://github.com/ultralytics/ultralytics/pull/27044) by [@raimbekovm](https://github.com/raimbekovm) captures exports on CPU to fix incorrect MPS output and CUDA crashes.

## 🖼 Visualization and Utility Fixes

- **Correct binary semantic-mask rendering:** [PR #27025](https://github.com/ultralytics/ultralytics/pull/27025) by [@raimbekovm](https://github.com/raimbekovm) plots only the foreground.
- **PIL text-coordinate support:** [PR #27038](https://github.com/ultralytics/ultralytics/pull/27038) by [@MohammadHijjawi97](https://github.com/MohammadHijjawi97) enables tuple coordinates in `Annotator.text` with PIL fonts.
- **Mask-IoU dtype handling:** [PR #27039](https://github.com/ultralytics/ultralytics/pull/27039) by [@cainiao33](https://github.com/cainiao33) promotes non-floating-point masks to `float32`.
- **Screenshot runner support:** [PR #27030](https://github.com/ultralytics/ultralytics/pull/27030) by [@glenn-jocher](https://github.com/glenn-jocher) adds Chromium system libraries for Playwright screenshots in the CPU runner image.

## 📚 Dataset Workflow Updates

- **Argoverse through Ultralytics Platform:** [PR #27033](https://github.com/ultralytics/ultralytics/pull/27033) by [@onuralpszr](https://github.com/onuralpszr) replaces Google Drive downloads with a Platform-based download and conversion workflow.
- **Clearer dataset subset documentation:** [PR #27029](https://github.com/ultralytics/ultralytics/pull/27029) by [@JaypeeLan](https://github.com/JaypeeLan) explains `fraction` subset selection for image-folder datasets. 🎉 Welcome, and thank you for your first contribution!

Documentation also now lists NXP among the available export options.

## ⚡ Try v8.4.203

Upgrade with:

```bash
pip install --upgrade ultralytics==8.4.203

```

Explore the [v8.4.203 release notes](https://github.com/ultralytics/ultralytics/releases/tag/v8.4.203) for release details, or review the [full changelog](https://github.com/ultralytics/ultralytics/compare/v8.4.202...v8.4.203) to see every change.

**Give the release a try and share your feedback below!** Whether you’re testing NXP export or updating an existing workflow, we’d love to hear what works well and what needs attention. 🙌