# New Release: Ultralytics v8.4.176

**URL:** <https://community.ultralytics.com/t/new-release-ultralytics-v8-4-176/2258>\
**Category:** Discussion\
**Tags:** ultralytics-official, releases, announcements\
**Created:** [October 13, 2026, 1:19pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-176/2258 "2026-10-13T13:19:05Z")\
**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:** [October 13, 2026, 1:19pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-176/2258/1 "2026-10-13T13:19:05Z")

</div>

# 🚀 Ultralytics v8.4.176: Smaller YOLO-World Checkpoints and More Reliable Workflows

**Ultralytics YOLO v8.4.176** brings dramatically smaller YOLO-World checkpoints and reliability improvements across validation, fine-tuning, segmentation, tracking, and deployment. Explore the [v8.4.176 release notes](https://github.com/ultralytics/ultralytics/releases/tag/v8.4.176) for the complete update.

## 🌟 Highlight: YOLO-World Checkpoints Shrink from 328.5 MB to 25.8 MB

Setting custom classes no longer caches the CLIP text encoder by default. For `yolov8s-worldv2.pt`, this reduced a saved checkpoint from **328.5 MB to 25.8 MB** , with **identical predictions after reload**. Explicit caching remains available when needed.

This makes custom-vocabulary checkpoints easier to store and share. See [PR #26647](https://github.com/ultralytics/ultralytics/pull/26647) by [@raimbekovm](https://github.com/raimbekovm) for the implementation.

## 🎯 Validation and Training Improvements

- **More reliable detection matching:** Predictions now match eligible targets in confidence order, reducing missed matches when alternate targets are available. [PR #26616](https://github.com/ultralytics/ultralytics/pull/26616) by [@Nikhi00718](https://github.com/Nikhi00718) implements this fix.
- **Safer YOLOE fine-tuning and resume:** Trained classification heads are preserved instead of reinitialized when resuming or re-fine-tuning positional-embedding checkpoints. [PR #26646](https://github.com/ultralytics/ultralytics/pull/26646) by [@raimbekovm](https://github.com/raimbekovm) preserves the fused head.
- **Worker-specific augmentation seeding:** Albumentations is reseeded in each DataLoader worker through [PR #26635](https://github.com/ultralytics/ultralytics/pull/26635) by [@8rulerstar](https://github.com/8rulerstar).

## 🖼 Segmentation and SAM Fixes

- **MobileSAM FP16 masks:** Native `LayerNorm2d` computation fixes empty masks during FP16 inference. [PR #26639](https://github.com/ultralytics/ultralytics/pull/26639) by [@raimbekovm](https://github.com/raimbekovm) delivers the fix.
- **Fresh SAM state for every source:** Reused predictors now reset state for new images and videos, thanks to [PR #26644](https://github.com/ultralytics/ultralytics/pull/26644) by [@raimbekovm](https://github.com/raimbekovm).
- **Segmentation regression reverted:** [PR #26416](https://github.com/ultralytics/ultralytics/pull/26416) by [@Nicholas022400701](https://github.com/Nicholas022400701) introduced mask thresholding and video-duration handling with `vid_stride`. The mask-thresholding portion was subsequently reverted in [PR #26645](https://github.com/ultralytics/ultralytics/pull/26645) by [@Laughing-q](https://github.com/Laughing-q) after it reduced YOLO26 segmentation mAP.

## ⚙ Deployment and Input Handling

- **ONNX GPU inference:** Custom input names and strided tensors are handled correctly by [PR #26618](https://github.com/ultralytics/ultralytics/pull/26618) from [@aswanth-07](https://github.com/aswanth-07).
- **OpenVINO NPU inference:** Batch and depth-output handling are improved in [PR #26641](https://github.com/ultralytics/ultralytics/pull/26641) by [@onuralpszr](https://github.com/onuralpszr).
- **Grayscale model inputs:** Floating-point NumPy color images are now accepted through [PR #26632](https://github.com/ultralytics/ultralytics/pull/26632) by [@aswanth-07](https://github.com/aswanth-07).
- **TensorRT 11 export dependencies:** Automatic installation of `nvidia-modelopt` now pins `torch`, thanks to [PR #26624](https://github.com/ultralytics/ultralytics/pull/26624) by [@Zephyr4772](https://github.com/Zephyr4772).

## 🎬 Video, Tracking, and Workflow Fixes

- **Accurate frame rates:** Saved videos preserve fractional FPS, and stream loaders handle high-FPS inputs correctly. [PR #26627](https://github.com/ultralytics/ultralytics/pull/26627) by [@Vaishnavi220506](https://github.com/Vaishnavi220506) improves both workflows.
- **Picklable tracker counters:** Per-tracker ID counters support copying and pickling on Python 3.14 through [PR #26638](https://github.com/ultralytics/ultralytics/pull/26638) by [@raimbekovm](https://github.com/raimbekovm).
- **Correct ReID crop colors:** [PR #26628](https://github.com/ultralytics/ultralytics/pull/26628) by [@cainiao33](https://github.com/cainiao33) fixes channel swapping in crops passed to embedding models.
- **Smoother benchmarking:** Legacy precision flags, model selection, and repeated directory runs are fixed in [PR #26640](https://github.com/ultralytics/ultralytics/pull/26640) by [@raimbekovm](https://github.com/raimbekovm).
- **Readable Windows logs:** Non-ASCII class names and paths are preserved by [PR #26626](https://github.com/ultralytics/ultralytics/pull/26626) from [@Nikhi00718](https://github.com/Nikhi00718).

## 📚 Documentation and Platform Updates

- **Updated workflow guidance:** Documentation clarifies precision, preprocessing, inference, and fine-tuning behavior. Ultralytics Platform documentation also covers relevance-based dataset sorting, email changes, and AI-assisted model comparison. [PR #26633](https://github.com/ultralytics/ultralytics/pull/26633) by [@raimbekovm](https://github.com/raimbekovm) brings these updates together.
- **Hailo-8 benchmark visualization:** [PR #26615](https://github.com/ultralytics/ultralytics/pull/26615) by [@lakshanthad](https://github.com/lakshanthad) adds a benchmark image to the documentation.

## 🙌 Community Thanks

Thank you to the YOLO community and Ultralytics team for the fixes, testing, and documentation improvements! A special welcome to **[@Zephyr4772](https://github.com/Zephyr4772)**, whose first contribution is included in this release.

## 🧪 Try It and Share Your Feedback

Upgrade to this release:

```bash
pip install --upgrade ultralytics==8.4.176

```

Try your training, inference, or export workflow and let us know how it goes. If something looks unexpected, share your environment, a minimal reproduction, and any relevant logs in the discussion.

You can review every change in the [full v8.4.175 → v8.4.176 changelog](https://github.com/ultralytics/ultralytics/compare/v8.4.175...v8.4.176) before upgrading.