# New Release: Ultralytics v8.4.194

**URL:** <https://community.ultralytics.com/t/new-release-ultralytics-v8-4-194/2289>\
**Category:** Discussion\
**Tags:** releases, announcements, ultralytics-official\
**Created:** [October 31, 2026, 1:05pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-194/2289 "2026-10-31T13:05:46Z")\
**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 31, 2026, 1:05pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-194/2289/1 "2026-10-31T13:05:46Z")

</div>

# 🚀 Ultralytics v8.4.194: More Reliable AutoBatch and Smoother Workflows

**Ultralytics YOLO v8.4.194 is here!** This release improves automatic batch sizing across tasks and delivers practical fixes for model loading, prediction saving, datasets, and deployment workflows. No new model architecture is introduced—this update focuses on reliability.

## 🌟 Key Improvements

- **Dataset-aware AutoBatch across all tasks.** Automatic batch sizing now uses the training dataset, accounting for dataset size and labels across classification, detection, and multi-dataset training. Backward-pass out-of-memory errors are treated as limits rather than valid measurements. In [PR #26956](https://github.com/ultralytics/ultralytics/pull/26956), [@raimbekovm](https://github.com/ultralytics/ultralytics/pull/26956) improved batch selection and addressed label-access failures in multi-dataset World/YOLOE training.

- **Safer, consistent model YAML parsing.** Activations must resolve to supported `torch.nn` modules, and model entries must be PyTorch module classes. In [PR #26963](https://github.com/ultralytics/ultralytics/pull/26963), [@Y-T-G](https://github.com/ultralytics/ultralytics/pull/26963) unified the parsing path. **Nonstandard custom YAML expressions may need updating.**

- **More robust prediction saving.** `Results.save()` and `plot(save=True)` now generate sensible `.jpg` filenames for webcams, videos, missing paths, and extensionless sources, creating parent directories when needed. In [PR #26958](https://github.com/ultralytics/ultralytics/pull/26958), [@Rahulbiradar9](https://github.com/ultralytics/ultralytics/pull/26958) addressed these common saving issues.

## 🛠 Dataset and Workflow Fixes

- **Fresh similarity-search indexes.** In [PR #26965](https://github.com/ultralytics/ultralytics/pull/26965), [@raimbekovm](https://github.com/ultralytics/ultralytics/pull/26965) made the index rebuild from the selected image directory on every startup. This avoids stale results from another dataset, with the trade-off of rebuilding embeddings each time.

- **Correct semantic-mask cache handling.** In [PR #26954](https://github.com/ultralytics/ultralytics/pull/26954), [@raimbekovm](https://github.com/ultralytics/ultralytics/pull/26954) ensured non-PNG mask paths are resolved before label-cache hashing.

- **Reliable classification dataset URLs.** In [PR #26879](https://github.com/ultralytics/ultralytics/pull/26879), [@Vaishnavi220506](https://github.com/ultralytics/ultralytics/pull/26879) fixed classification dataset URL-scheme handling.

- **Case-insensitive URL routing.** In [PR #26964](https://github.com/ultralytics/ultralytics/pull/26964), [@cainiao33](https://github.com/ultralytics/ultralytics/pull/26964) updated remaining routing sites to handle schemes such as `HTTPS://`, `GS://`, and `UL://` without case sensitivity.

- **Independent Colab settings IDs.** In [PR #26959](https://github.com/ultralytics/ultralytics/pull/26959), [@glenn-jocher](https://github.com/ultralytics/ultralytics/pull/26959) gave each Colab VM its own anonymous settings ID.

## 📊 Benchmarking and Documentation

- **Broader GraphDef benchmarking.** In [PR #26952](https://github.com/ultralytics/ultralytics/pull/26952), [@glenn-jocher](https://github.com/ultralytics/ultralytics/pull/26952) added TensorFlow GraphDef benchmark coverage for OBB and pose.

- **Clearer per-class metric interpretation.** In [PR #26955](https://github.com/ultralytics/ultralytics/pull/26955), [@Echoslayer](https://github.com/ultralytics/ultralytics/pull/26955) clarified that `metrics.box.maps` assigns the overall mAP to classes absent from validation—not a measured per-class score.

- **Updated COCO JSON cache guidance.** In [PR #26960](https://github.com/ultralytics/ultralytics/pull/26960), [@Echoslayer](https://github.com/ultralytics/ultralytics/pull/26960) updated the guide to reflect that cache hashing includes modification time.

- **Refreshed integration pages.** In [PR #26957](https://github.com/ultralytics/ultralytics/pull/26957), [@raimbekovm](https://github.com/ultralytics/ultralytics/pull/26957) corrected stale framing, logger defaults, and navigation order.

- **Cleaner settings help.** In [PR #26962](https://github.com/ultralytics/ultralytics/pull/26962), [@glenn-jocher](https://github.com/ultralytics/ultralytics/pull/26962) fixed help URL punctuation.

## ⚡ Try v8.4.194

Install this release with:

```bash
pip install --upgrade ultralytics==8.4.194

```

Explore the [v8.4.194 release notes](https://github.com/ultralytics/ultralytics/releases/tag/v8.4.194) for the release overview, or review the [full changelog](https://github.com/ultralytics/ultralytics/compare/v8.4.193...v8.4.194) for all changes.

**How does this update perform in your workflow?** Share your AutoBatch results, saving improvements, or any issues in the replies. Thanks to everyone who contributed! 🙌