# New Release: Ultralytics v8.4.38

**URL:** <https://community.ultralytics.com/t/new-release-ultralytics-v8-4-38/1936>\
**Category:** Discussion\
**Tags:** releases, announcements, ultralytics-official\
**Created:** [April 16, 2026, 1:35pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-38/1936 "2026-04-16T13:35:30Z")\
**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:** [April 16, 2026, 1:35pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-38/1936/1 "2026-04-16T13:35:30Z")

</div>

## Ultralytics v8.4.38 is out 🚀

**Summary:** Ultralytics **v8.4.38** is a stability-focused release centered on **more reliable and consistent model export behavior** across deployment formats, plus important fixes for training, tracking, and SAM3 workflows. If you export often, train across multiple GPUs, or rely on tracking and prompting features, this update is definitely worth pulling. 🙌

You can explore the full release in the [v8.4.38 release notes](https://github.com/ultralytics/ultralytics/releases/tag/v8.4.38) and see every commit in the [full changelog](https://github.com/ultralytics/ultralytics/compare/v8.4.37...v8.4.38).

* * *

## Highlights ✨

### More reliable export across deployment targets 📦

The biggest change in `v8.4.38` is the export cleanup and standardization in [PR #24120 by @Laughing-q](https://github.com/ultralytics/ultralytics/pull/24120), with profile details on [@Laughing-q’s GitHub](https://github.com/Laughing-q).

This work unifies standalone export behavior across:

- CoreML
- ONNX
- OpenVINO
- TensorFlow
- TorchScript
- ExecuTorch
- Axelera
- RKNN
- IMX
- NCNN
- MNN
- Paddle

Key benefits include:

- CoreML now uses the model’s real input name instead of assuming `"image"`
- clearer standardized argument names like `output_file` and `output_dir`
- better handling for multi-input export cases, especially in ONNX and OpenVINO
- more robust path handling in TensorFlow and IMX export flows

For deployment teams using **Ultralytics YOLO** , this should make exports more predictable and much less fragile across runtimes and devices. ✅

### Training reliability improvements 🏋

This release also addresses a few important training edge cases:

- Resume fixes for non-end2end models in [PR #24173 by @fcakyon](https://github.com/ultralytics/ultralytics/pull/24173), from [@fcakyon](https://github.com/fcakyon)
- Safer DDP stride setup in [PR #24208 by @Laughing-q](https://github.com/ultralytics/ultralytics/pull/24208), from [@Laughing-q](https://github.com/Laughing-q)
- Sentry-related robustness fixes in [PR #24220 by @glenn-jocher](https://github.com/ultralytics/ultralytics/pull/24220), from [@glenn-jocher](https://github.com/glenn-jocher)

These updates help reduce resume, checkpoint, and multi-GPU surprises in production training pipelines. ⚙

### Tracking now behaves more intuitively 🎯

`track_buffer` now consistently works as a true **frame count** , without hidden FPS scaling, thanks to [PR #24247 by @TimSchoonbeek](https://github.com/ultralytics/ultralytics/pull/24247), from [@TimSchoonbeek](https://github.com/TimSchoonbeek).

That means tracking behavior should now better match exactly what you configure, especially for videos that are not 30 FPS. 🎥

### SAM3 fixes for stability and prompting 🧠

Two useful SAM3 fixes landed in this release:

- Presence-logit clamp fix in [PR #24213 by @Y-T-G](https://github.com/ultralytics/ultralytics/pull/24213), from [@Y-T-G](https://github.com/Y-T-G)
- Text-only prompt behavior fix in [PR #24244 by @Y-T-G](https://github.com/ultralytics/ultralytics/pull/24244), also from [@Y-T-G](https://github.com/Y-T-G)

Together these improve numerical behavior and make text-only grounding more consistent. 🔬

* * *

## Improvements 🔧

### Export robustness beyond the main refactor

A few additional export-related fixes help round out this release:

- YOLOE export now skips incompatible fusion when `lrpc` is present in [PR #24239 by @Y-T-G](https://github.com/ultralytics/ultralytics/pull/24239), from [@Y-T-G](https://github.com/Y-T-G)
- `fuse()` now returns the model instance in [PR #24246 by @Y-T-G](https://github.com/ultralytics/ultralytics/pull/24246), from [@Y-T-G](https://github.com/Y-T-G)
- Axelera install command updated for prerelease packages in [PR #24190 by @onuralpszr](https://github.com/ultralytics/ultralytics/pull/24190), from [@onuralpszr](https://github.com/onuralpszr)
- `axelera-runtime` prerelease install support in [PR #24230 by @lakshanthad](https://github.com/ultralytics/ultralytics/pull/24230), from [@lakshanthad](https://github.com/lakshanthad)
- OpenVINO 2026 Conda CI segfault fix in [PR #24224 by @onuralpszr](https://github.com/ultralytics/ultralytics/pull/24224), from [@onuralpszr](https://github.com/onuralpszr)
- JetPack 5 Docker `torch` and `torchvision` version fix in [PR #24199 by @lakshanthad](https://github.com/ultralytics/ultralytics/pull/24199), from [@lakshanthad](https://github.com/lakshanthad)

### Test coverage and internal quality

This release also strengthens test reliability and coverage:

- OBB and Pose test coverage added in [PR #24197 by @Laughing-q](https://github.com/ultralytics/ultralytics/pull/24197), from [@Laughing-q](https://github.com/Laughing-q)
- Solution tests updated to use cached session assets in [PR #24237 by @Laughing-q](https://github.com/ultralytics/ultralytics/pull/24237), from [@Laughing-q](https://github.com/Laughing-q)

* * *

## Docs updates 📚

Several documentation improvements shipped alongside the code fixes:

- [PR #24194 by @raimbekovm](https://github.com/ultralytics/ultralytics/pull/24194) from [@raimbekovm](https://github.com/raimbekovm) adds `trackzone` to the solutions CLI enum
- [PR #24195 by @raimbekovm](https://github.com/ultralytics/ultralytics/pull/24195) from [@raimbekovm](https://github.com/raimbekovm) fixes docs macro type/default drift
- [PR #24189 by @raimbekovm](https://github.com/ultralytics/ultralytics/pull/24189) from [@raimbekovm](https://github.com/raimbekovm) fixes export-table drift for ExecuTorch, Axelera, and IMX
- [PR #24209 by @raimbekovm](https://github.com/ultralytics/ultralytics/pull/24209) from [@raimbekovm](https://github.com/raimbekovm) fixes `solutions-args` colormap default drift
- [PR #24207 by @raimbekovm](https://github.com/ultralytics/ultralytics/pull/24207) from [@raimbekovm](https://github.com/raimbekovm) adds missing validated args to TorchScript, CoreML, and ExecuTorch docs
- [PR #24211 by @raimbekovm](https://github.com/ultralytics/ultralytics/pull/24211) from [@raimbekovm](https://github.com/raimbekovm) clarifies that YOLO12 pretrained weights are detect-only
- [PR #24234 by @raimbekovm](https://github.com/ultralytics/ultralytics/pull/24234) from [@raimbekovm](https://github.com/raimbekovm) clarifies TT100K categories vs trainable classes
- [PR #24217 by @raimbekovm](https://github.com/ultralytics/ultralytics/pull/24217) from [@raimbekovm](https://github.com/raimbekovm) adds CLI tabs for all 12 YOLO solutions
- [PR #24214 by @laodouya](https://github.com/ultralytics/ultralytics/pull/24214) from [@laodouya](https://github.com/laodouya) updates annotation viewer zoom shortcut docs
- [PR #24221 by @onuralpszr](https://github.com/ultralytics/ultralytics/pull/24221) from [@onuralpszr](https://github.com/onuralpszr) cleans up security docs and outdated links
- [PR #24232 by @Laughing-q](https://github.com/ultralytics/ultralytics/pull/24232) from [@Laughing-q](https://github.com/Laughing-q) fixes table formatting in `annotation.md`

* * *

## Why this matters 💡

This is not a flashy feature release, but it is a very practical one:

- **Export users** get more consistent outputs across runtimes
- **Training users** get fewer edge-case failures when resuming or using DDP
- **Tracking users** get `track_buffer` behavior that matches configuration
- **SAM3 users** get more stable and accurate prompt handling

In short: **v8.4.38 makes existing workflows safer, clearer, and more production-friendly**. 🙌

* * *

## New contributor 🌟

A big welcome to [@TimSchoonbeek](https://github.com/TimSchoonbeek), who made their first contribution with [PR #24247](https://github.com/ultralytics/ultralytics/pull/24247)! 🎉

* * *

## Try it out 🧪

If you want to upgrade locally, you can pull the latest release with:

```bash
pip install -U ultralytics

```

If you’re starting a new project, we recommend using [Ultralytics YOLO26 on Ultralytics Platform](https://platform.ultralytics.com/ultralytics/yolo26), which is our latest stable and recommended model family for all use cases. You can also explore the full [Ultralytics Platform documentation](https://docs.ultralytics.com/platform/) for annotation, training, deployment, and monitoring workflows.

* * *

## Feedback welcome 💬

Please give **v8.4.38** a try and let us know how it performs in your training, export, and deployment pipelines. Feedback, edge cases, and regression reports are always appreciated and help make YOLO better for everyone.