# New Release: Ultralytics v8.4.3

**URL:** <https://community.ultralytics.com/t/new-release-ultralytics-v8-4-3/1759>\
**Category:** Discussion\
**Tags:** ultralytics-official, releases, announcements\
**Created:** [January 16, 2026, 1:16pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-3/1759 "2026-01-16T13:16:41Z")\
**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:** [January 16, 2026, 1:16pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-3/1759/1 "2026-01-16T13:16:42Z")

</div>

# 📣 Ultralytics v8.4.3 is out! 🚀

**TL;DR:** Ultralytics `v8.4.3` speeds up **Ultralytics Platform NDJSON dataset conversion/downloads** ⚡, improves **training metric correctness & reliability** 🧠, and refreshes defaults/docs to make **YOLO26** the smoothest starting point 📚.

* * *

## 🌟 New Features

### 🚀 Faster NDJSON → YOLO dataset conversion (Ultralytics Platform data)

Big speed + robustness wins for Platform-backed datasets in **[PR #23257](https://github.com/ultralytics/ultralytics/pull/23257)** by **[@glenn-jocher](https://github.com/glenn-jocher)**:

- Lazy-loads `aiohttp` only when NDJSON conversion is used (faster startup, fewer unnecessary deps) 📦
- Simplifies async image download logic + improves concurrency scaling with dataset size ⚡
- More robust handling of Platform URL inputs 🌐
- Includes version bump `8.4.2 → 8.4.3` 🔖

### 🌐 Configurable Ultralytics Platform base URL

For teams testing staging/local setups, **[PR #23256](https://github.com/ultralytics/ultralytics/pull/23256)** by **[@glenn-jocher](https://github.com/glenn-jocher)** adds:

- `ULTRALYTICS_PLATFORM_URL` env var to point callbacks/links to custom Platform environments 🧪

* * *

## 📈 Improvements

### 🏷 Defaults and examples move to YOLO26

Docs/examples now default to `yolo26n.pt` in **[PR #23242](https://github.com/ultralytics/ultralytics/pull/23242)** by **[@Laughing-q](https://github.com/Laughing-q)**:

- `YOLO()` and CLI fallback model become `yolo26n.pt` ✅
- Helps new users start on the latest recommended model family ( **YOLO26** ) with fewer naming bumps

### ⚙ Optimizer + warmup logic is more reliable

Training behavior is more predictable in **[PR #23234](https://github.com/ultralytics/ultralytics/pull/23234)** by **[@Laughing-q](https://github.com/Laughing-q)**:

- Parameter groups explicitly labeled; warmup LR targets the **bias** group by _name_ (not position) 🔥
- “Auto” optimizer strategy simplified around MuSGD with improved defaults 🎛

### 🧩 IMX inference/export consistency

Better shape-handling and a simpler decode path in **[PR #23235](https://github.com/ultralytics/ultralytics/pull/23235)** by **[@Laughing-q](https://github.com/Laughing-q)**:

- More robust anchor/stride refresh for changing input shapes 🔧

### 📊 Benchmark tables clarified (end-to-end metrics)

README tables now clearly distinguish e2e evaluation metrics in **[PR #23238](https://github.com/ultralytics/ultralytics/pull/23238)** by **[@Laughing-q](https://github.com/Laughing-q)**.

### 🍓 Raspberry Pi 5 guide refreshed with YOLO26 benchmarks

Updated benchmarks + ExecuTorch results in **[PR #23227](https://github.com/ultralytics/ultralytics/pull/23227)** by **[@lakshanthad](https://github.com/lakshanthad)**.

* * *

## 🐛 Bug Fixes

### 🧍 Pose training logs are more accurate

Avoids reporting `rle_loss` when unsupported in **[PR #23230](https://github.com/ultralytics/ultralytics/pull/23230)** by **[@lmycross](https://github.com/lmycross)**:

- Cleaner logs, fewer confusing metrics 🧾

### ✅ Fix duplicated `Results.summary()` entries

Prevents duplicated rows in summaries in **[PR #23218](https://github.com/ultralytics/ultralytics/pull/23218)** by **[@xusuyong](https://github.com/xusuyong)**:

- Cleaner analytics/logging 🧹

* * *

## 🎉 New Contributors

Welcome **[@xusuyong](https://github.com/xusuyong)** — first contribution landed via **[PR #23218](https://github.com/ultralytics/ultralytics/pull/23218)**! 🙌

* * *

## 🧪 Try it now

Upgrade with:

```bash
pip install -U ultralytics

```

Quick sanity check with YOLO26:

```python
from ultralytics import YOLO

model = YOLO("yolo26n.pt")
model.predict("https://ultralytics.com/images/bus.jpg")

```

* * *

## 🔎 Changelog & Release

Browse the full release details in the **[v8.4.3 GitHub Release](https://github.com/ultralytics/ultralytics/releases/tag/v8.4.3)**, or review the complete diff in the **[v8.4.2 → v8.4.3 changelog comparison](https://github.com/ultralytics/ultralytics/compare/v8.4.2...v8.4.3)**.

* * *

## 💬 Feedback

Tried `v8.4.3` yet? Let us know what improved (or what didn’t) — especially if you’re using Ultralytics Platform datasets, training at scale, or exporting IMX. Your reports help us keep YOLO fast, reliable, and easy to use.