# New Release: Ultralytics v8.4.59

**URL:** <https://community.ultralytics.com/t/new-release-ultralytics-v8-4-59/2022>\
**Category:** Discussion\
**Tags:** ultralytics-official, releases, announcements\
**Created:** [June 1, 2026, 3:04pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-59/2022 "2026-06-01T15:04:46Z")\
**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:** [June 1, 2026, 3:04pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-59/2022/1 "2026-06-01T15:04:46Z")

</div>

# Ultralytics v8.4.59 is here 🚀

**Quick summary:** Ultralytics `v8.4.59` improves **Rockchip RKNN export** with safer, clearer, and easier **INT8 quantization** support. This release makes deployment on Rockchip edge devices more reliable, especially for INT8-only chips, while keeping the export workflow aligned with the familiar Ultralytics YOLO experience 🙌

## 🌟 Highlights

This release focuses on a major RKNN export upgrade:

- **INT8 RKNN export support is now available** with `int8=True`
- **Standard dataset YAMLs** passed with `data=...` are now supported directly for calibration
- **`fraction=...`** is supported to control calibration dataset usage
- **Calibration file validation** now happens before export proceeds
- **INT8-only Rockchip chips** like `RV1103`, `RV1106`, `RV1103B`, and `RV1106B` are better supported
- **RKNN docs and export tables** were updated for clarity
- **RKNN API error handling** is now more robust with clearer runtime failures

If you’re deploying YOLO models to Rockchip hardware, this is a great release to try.

## 🆕 New Features

### RKNN INT8 export support for Rockchip devices 📱

With `v8.4.59`, RKNN export now supports:

- `int8=True`
- `data=...`
- `fraction=...`

This enables **INT8-quantized RKNN model export** , which is especially important for Rockchip devices that require INT8 models for deployment.

Instead of managing RKNN-specific calibration files manually, users can continue using the standard Ultralytics workflow by passing a dataset YAML during export. Ultralytics handles the internal conversion to RKNN’s calibration file format automatically, keeping the process simpler and more consistent.

## ⚙ Improvements

### Safer calibration dataset handling ✅

A key improvement in this release is early validation of the internally generated RKNN calibration image list before export loads the RKNN Toolkit.

That means setup issues are caught earlier, with clearer error messages, instead of failing later in more confusing ways.

### Better support for INT8-only chips 🧠

This release expands support for Rockchip hardware that cannot use floating-point RKNN models. The exporter also warns users and auto-adjusts behavior when a selected target requires `int8=True`, helping reduce configuration mistakes.

### Clearer docs and export guidance 📚

RKNN documentation and export reference tables were updated to better explain:

- floating-point export behavior for FP16-capable targets
- when `int8=True` is required
- how `data` and `fraction` are used for calibration

This should make RKNN export easier to understand for both new and experienced users.

## 🐛 Bug Fixes

### Improved RKNN API error handling 🛡

Internal RKNN API calls now validate return values more carefully. Failures during config, ONNX loading, build, or export are surfaced as clearer runtime errors, making debugging much easier.

## 🔀 Pull Requests

Included in this release:

- **Rockchip RKNN INT8 export option** by [@glenn-jocher](https://github.com/glenn-jocher) in [PR #24656](https://github.com/ultralytics/ultralytics/pull/24656)
- **Validate RKNN INT8 calibration dataset path** by [@glenn-jocher](https://github.com/glenn-jocher) in [PR #24659](https://github.com/ultralytics/ultralytics/pull/24659)

You can also review the [full changelog for `v8.4.59`](https://github.com/ultralytics/ultralytics/compare/v8.4.58...v8.4.59) or browse the [release page for Ultralytics `v8.4.59`](https://github.com/ultralytics/ultralytics/releases/tag/v8.4.59).

## 💡 Why this matters

This update makes RKNN export:

- **more reliable** through earlier validation
- **easier to use** with standard dataset YAML inputs
- **more compatible** with INT8-only Rockchip NPUs
- **cleaner** by hiding RKNN-specific implementation details
- **safer** with clearer warnings and improved runtime checks

For edge deployment workflows, especially on constrained hardware, these improvements can make a real difference in usability and deployability.

## ▶ Try it out

Update with:

```bash
pip install -U ultralytics

```

Then export with a workflow like:

```python
from ultralytics import YOLO

model = YOLO("yolo11n.pt")
model.export(format="rknn", int8=True, data="coco8.yaml", fraction=0.5)

```

If you’re starting a new project, we recommend exploring [Ultralytics YOLO26 on the Ultralytics Platform](https://platform.ultralytics.com/ultralytics/yolo26), which is our latest stable and recommended model family for all use cases.

## 💬 Feedback

Please give `v8.4.59` a try and let us know how it works for your Rockchip deployment workflows. Feedback, testing results, and edge-case reports are always appreciated and help the whole Ultralytics community improve faster 🙏