What benefits does yolov26 offer over yolov11?

[Raj2032]

I have tested both yolov26 and yolov11 both using small models and accuracy is slightly worse on yolov26 than it is on yolov26 (I did a “proof of concept” with my training).

I was wondering is this the reason due to the trade off favoring slightly higher speeds over accuracy?

I used the exact same paramaters, for example:

dataset_custom.yaml

train: images/train
val: images/val

nc: 1

names: ["mc"]

train.py

from ultralytics import YOLO

# model = YOLO("yolo11s.pt")
model = YOLO("yolo26s.pt")
model.train(data = "dataset_custom.yaml", imgsz = 640,
batch = 8, epochs = 100, workers = 0, device = "cpu")

I am pretty sure this is optimal for both yolo11s and yolo26s

(I had 11 images for training and 5 images for validation, its just for a proof of concept rather than an official product I am trying to build, hence the lack of image training).

[UltralyticsAssistant]

👋 Hello @Raj2032, thank you for your interest in Ultralytics 🚀! We recommend a visit to the Docs for new users where you can find many Python and CLI usage examples and where many of the most common questions may already be answered.

This is an automated response 🤖—an Ultralytics engineer will also assist soon 🙏

If this is a 🐛 Bug Report, please provide a minimum reproducible example to help us debug it. In your case, an MRE would ideally include the exact commands/code you ran, the full console logs, the ultralytics version, and enough detail for someone to reproduce the YOLO11 vs YOLO26 comparison on the same data/settings.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results. With very small proof-of-concept datasets (like 11 train / 5 val), sharing training curves/logs and a couple of representative sample images can be especially helpful 📌

Join the Ultralytics community where it suits you best. For real-time chat, head to Discord 🎧. Prefer in-depth discussions? Check out Discourse. Or dive into threads on our Subreddit to share knowledge with the community.

Upgrade

Upgrade to the latest ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8 to verify your issue is not already resolved in the latest version:

pip install -U ultralytics

Environments

YOLO may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all Ultralytics CI tests are currently passing. CI tests verify correct operation of all YOLO Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

[glenn-jocher]

With only 11 train / 5 val images, the reported mAP/accuracy will have huge variance (and can flip between runs), so a small “YOLO11s > YOLO26s” result is usually noise/overfit rather than an intentional speed-for-accuracy tradeoff; YOLO26’s main gains are simpler end-to-end deployment (NMS-free) and faster CPU inference, as summarized in the YOLO11 vs YOLO26 FAQ. To make the comparison meaningful, please rerun a few times with a fixed seed and share the full training logs plus yolo checks output (and if possible a small sample of images/labels), and we can help sanity-check whether you’re just hitting small-data instability or a real regression on your setup.

[Raj2032]

Hi mate, when you say to rerun a few times do you mean I should rerun training a few times or rerun the model a few times?

When you say with a fixed seed is this something that should go inside train.py? What value should I set it to?

[glenn-jocher]

I meant rerun the training + validation a few times (not inference), since training is stochastic; set the same seed inside model.train() for both models (any integer is fine—0 or 42 are common) and compare the resulting best.pt metrics across runs—see the Train mode arguments for details.

from ultralytics import YOLO

model = YOLO("yolo26s.pt")  # or "yolo11s.pt"
model.train(data="dataset_custom.yaml", imgsz=640, batch=8, epochs=100, workers=0, device="cpu", seed=0)

[Y-T-G]

You should use a larger dataset for more representative results because you are trying to learn millions of parameters from just 11 images.